Add semantic contexts for Swift, Kotlin, PHP, Rust, and Go

[TheFedaikin]

Summary

Second wave of the cross-language semantic extractor (after #996). Brings the same vocabulary to Swift,
Kotlin, PHP, Rust, and Go: normalized type-relationship edges and references edges with
parameter_type / return_type / generic_arg / field / attribute contexts.

Storage is unchanged — graph.json simply carries more source-backed edges.

What changed

• Swift. Pre-scan protocol_declaration vs class/struct/enum/actor declarations to classify each
  inheritance_specifier as inherits (class supertype) or implements (protocol / extension
  conformance). protocol P: Q emits inherits. Method parameter and return_type, plus
  property_declaration annotations, emit references with the standard contexts.
• Kotlin. Split delegation_specifier into inherits (constructor_invocation) vs implements
  (plain user_type). Function function_value_parameters and post-colon return types emit references;
  property_declaration types emit field.
• PHP. base_clause → inherits, class_interface_clause → implements, use Trait →
  mixes_in. Typed properties, simple parameters, and post-) return types emit references. Also fixes
  the event-listener property_declaration handler to only short-circuit when it actually matched a
  $listen / $subscribe property — non-listener typed properties were previously dropped.
• Rust. impl Trait for T emits T implements Trait. Supertraits on trait_item via
  trait_bounds emit inherits. Struct field_declaration types emit references field;
  function_item parameters and return types emit parameter_type / return_type, with generic
  arguments emitted as generic_arg.
• Go. Struct field_declaration without a field_identifier (the Go embedded-field syntax) emits
  embeds; named fields emit references field. Interface type_elem entries emit embeds (interface
  embedding). method_declaration / function_declaration parameters and result (handling both
  single-type and parameter_list forms) emit parameter_type / return_type.

Fixture changes

• tests/fixtures/sample.swift: adds BaseProcessor class, Result<T>, a current: Result<DataProcessor> property, and a run(input: DataProcessor) -> Result<DataProcessor> method.
• tests/fixtures/sample.kt: adds Loggable interface, BaseProcessor open class, Result<T>, a
  DataProcessor : BaseProcessor(), Loggable that overrides log(), a current: Result<DataProcessor>
  property, and a run method.
• tests/fixtures/sample.php: adds Loggable interface, HasName trait, BaseProcessor, Result,
  and a DataProcessor extends BaseProcessor implements Loggable with use HasName;, a private Result $current; field, and a run(DataProcessor $input): Result method.
• tests/fixtures/sample.rs: adds Processor / Logger: Processor traits, Result<T>, a
  DataProcessor struct with a Result<DataProcessor> field, an impl Processor for DataProcessor, and
  a build(input: DataProcessor) -> Result<DataProcessor> method.
• tests/fixtures/sample.go: adds Logger / Reader interfaces, a ReaderLogger that embeds both,
  BaseProcessor, Result, and a DataProcessor that embeds BaseProcessor with a *Result field plus
  a Build(input *DataProcessor) (*Result, error) method.

These fixture additions keep all pre-existing per-language assertions intact. The Swift
test_swift_conformance_edge / test_swift_extension_conformance_edge tests are renamed to
test_swift_protocol_conformance_emits_implements / test_swift_extension_conformance_emits_implements
and now assert implements to match the new split.

Tests

Full pytest suite: 1294 passed, 11 skipped.

New tests:

• test_swift_protocol_conformance_emits_implements
• test_swift_extension_conformance_emits_implements
• test_swift_splits_inherits_and_implements
• test_swift_parameter_return_generic_and_field_contexts
• test_kotlin_splits_inherits_and_implements
• test_kotlin_parameter_return_generic_and_field_contexts
• test_php_splits_inherits_implements_mixes_in
• test_php_property_parameter_and_return_contexts
• test_rust_trait_impl_emits_implements
• test_rust_supertrait_emits_inherits
• test_rust_struct_field_emits_field_context
• test_rust_method_parameter_return_and_generic_contexts
• test_go_embeds_struct_field
• test_go_interface_embedding_emits_embeds
• test_go_struct_named_field_emits_field_context
• test_go_method_parameter_return_contexts

Compatibility note

Swift inherits edges for protocol conformance are now implements. This matches the Java migration
applied in #996 and the C# split that also landed in PR 1 — protocols / interfaces / traits are
uniformly modeled as implements, class supertyping as inherits. Existing call/contains/method/import
edges and node IDs are unchanged.

Non-goals

• PR 3 (ObjC, Julia, C/C++, Scala, Dart, PowerShell, Fortran, SystemVerilog) is the planned follow-up.

[safishamsi]

Thanks for the PR — the design is solid and the test coverage is strong across all five languages. One issue before merge: there's a latent crash in the Go method_declaration block where emit_go_method_refs(...) appears to be outside the if name_node: guard. If name_node is None, method_nid will be undefined and you'll get an UnboundLocalError. Tests pass currently only because fixtures always have a name node. Could you fix the indentation and open a fresh PR against v8?

[safishamsi]

Thanks for the PR — the design is solid and the test coverage is strong across all five languages. One issue before merge: there's a latent crash in the Go method_declaration block where emit_go_method_refs(...) appears to be outside the if name_node: guard. If name_node is None, method_nid will be undefined and you'll get an UnboundLocalError. Tests pass currently only because fixtures always have a name node. Could you fix the indentation and open a fresh PR against v8?

[TheFedaikin]

@safishamsi I redid the test and indent looked right. Re-pushed the changes with a proof. I think it's getting really unwieldy with the current line-count for both manual and AI-driven reviews. Can I do some refactoring of the methods so they are more sustainable in this? I was hesitant of doing this in the middle of adding the functionality to unblock some of the stuff I needed from graphify but I get a formal approval I can be more confident with the time invested so it benefits all of the contributors.

[safishamsi]

Fetched the branch and checked line 4719 directly:

```
4716: method_nid = _make_id(stem, method_name) # 20 spaces — inside if name_node:
4717: add_node(method_nid, ...) # 20 spaces
4718: add_edge(file_nid, method_nid, ...) # 20 spaces
4719: emit_go_method_refs(node, method_nid, line) # 16 spaces — still OUTSIDE
```

The bug is still there. The structural test missed it because it inspects the AST of the source file but the reference to method_nid at line 4719 reads as an undefined name when name_node is None — the AST-level checker didn't catch the indent difference.

The fix is a single 4-space indent on line 4719:

```python
emit_go_method_refs(node, method_nid, line) # move inside if name_node:
```

On the refactoring question — once this is fixed and merged, a follow-up PR for internal refactoring of the Go extractor methods is welcome, just keep it separate from the new functionality.

[TheFedaikin]

When I changed the indent like you mentioned I saw the errors like so:

  ⎿    ✘ [Line 3875:16] Import "tree_sitter_sql" could not be resolved [reportMissingImports]
     (Pyright)
  ⎿    ✘ [Line 7799:16] Import "tree_sitter_pascal" could not be resolved
     [reportMissingImports] (Pyright)
  ⎿    ✘ [Line 3975:45] "lower" is not a known attribute of "None" [reportOptionalMemberAccess]
      (Pyright)
  ⎿    ✘ [Line 3976:71] "lower" is not a known attribute of "None" [reportOptionalMemberAccess]
      (Pyright)
  ⎿    ✘ [Line 3978:60] "lower" is not a known attribute of "None" [reportOptionalMemberAccess]
      (Pyright)
  ⎿    ✘ [Line 3976:98] Argument of type "str | None" cannot be assigned to parameter "parts"
     of type "str" in function "_make_id"
       Type "str | None" is not assignable to type "str"
         "None" is not assignable to "str" [reportArgumentType] (Pyright)
  ⎿    ✘ [Line 4352:22] Cannot access attribute "type" for class "object"
       Attribute "type" is unknown [reportAttributeAccessIssue] (Pyright)
  ⎿    ✘ [Line 4353:36] Cannot access attribute "children" for class "object"
       Attribute "children" is unknown [reportAttributeAccessIssue] (Pyright)
  ⎿    ✘ [Line 4559:32] Cannot access attribute "children" for class "object"
       Attribute "children" is unknown [reportAttributeAccessIssue] (Pyright)
  ⎿    ✘ [Line 6291:21] Cannot access attribute "type" for class "object"
       Attribute "type" is unknown [reportAttributeAccessIssue] (Pyright)

So it's definitely wasn't a fix here.
I think reading this was a chore so I refactored it a little - current change reads better and should be correct. Feel free to re-review!