Move object converters into a sibling FfiConverter class

[1egoman]

FIxes #155

Summary

ffi_converter_name returned the canonical name of the object for ObjectImpl::Struct and ObjectImpl::Trait. The converter statics therefore landed on the object class itself, where a Rust method of the same name collided with them.

The function now returns FfiConverter<CanonicalName> for both variants, and generate_object and generate_trait_object emit that class as a sibling of the object. Records, enums and callback interfaces already follow this pattern.

Effect on generated code

 class Recorder implements RecorderInterface {
-  static Pointer<Void> lower(Recorder value) { /* ... */ }
-  static int allocationSize(Recorder value) { /* ... */ }
-  static LiftRetVal<Recorder> read(Uint8List buf) { /* ... */ }
-  static int write(Recorder value, Uint8List buf) { /* ... */ }
-
   void write({required Uint8List chunk}) { /* ... */ }
 }
+
+class FfiConverterRecorder {
+  static Recorder lift(Pointer<Void> ptr) { /* ... */ }
+  static Pointer<Void> lower(Recorder value) { /* ... */ }
+  static int allocationSize(Recorder value) { /* ... */ }
+  static LiftRetVal<Recorder> read(Uint8List buf) { /* ... */ }
+  static int write(Recorder value, Uint8List buf) { /* ... */ }
+}

Tests

I opted to not add any explicit tests for this since the existing test suite covers all these paths quite well - it's exercising all these existing internal bindgen method calls. If you'd like though, I could add something which goes through the whole call cycle with a method named write() on an object just to validate this 100% and protect against a regression.

[1egoman]

Important callout that this changes the generated API - Recorder.lower(x) becomes FfiConverterRecorder.lower(x). Code that a consumer writes by hand I wouldn't think normally call these statics, because they belong to the FFI layer. But if any code does, it would need to change.

[1egoman]

Note that the error handler now points at the converter - an [Error] interface emits an error handler class, and that handler called Recorder.read. It now calls FfiConverterRecorder.read.

[1egoman]

@chavic For reference, the two test failures on here seem to be the ones which we discussed on #150 (the lint failure, and the flaky dart_async test which I will leave unaddressed per #139)

Would you be willing to review this in its current state, or would you like to see the linting issue get resolved first? I think other than the linting issue this should be ready for review from my end.

[1egoman]

Rebased on top of latest main to fix the lint errors.

[chavic]

Nice fix, thank you. On your API callout: the rename is safe to take. The generator emits these calls from one source, so no stale call site remains, and the hand-written Dart in payjoin-ffi and bdk-dart calls none of these statics.

Please add a small fixture with an object that has a method named read or lower. That is the collision this change corrects, and no test guards it yet.