fix(#6028): capture the full Solidity declaration signature, not just the name

[arthurgeron]

declSignature scans forward from the declaration regex match with paren depth and stops at the first { or ; at depth zero, so the parameter list and any returns (...) clause are captured. The extractor previously stored the regex match span itself, which ends at the opening paren.

• It does not repeat the comment and string handling in extractBracedBody. findContracts is fed stripCommentsAndStrings output, so a { or ; inside a comment or a literal cannot end a signature early.
• The literal's contents are blanked too, so onlyRole("ADMIN)ROLE") is captured as onlyRole( ). The third test pins that rather than leaving it to be found in review.
• A declaration with neither terminator falls back to the previous span, so truncated source keeps today's behaviour instead of running to the end of the contract.

Tests cover the multi-line, nested-paren and ;-terminated interface shapes, events, modifiers, and the comment/string case. Each fails against v0.1.9.

go test ./... green (224 packages), go vet and gofmt -l internal/extractors/solidity clean, and tools/coverage validate, backfill --check and fmt --check exit 0 with gen leaving docs/coverage byte-identical. Coverage matrix updated per AGENTS.md.

Closes #6028

[cajasmota]

Thank you for this — it's a genuinely nice piece of work, and a few things stand out.

The diagnosis is exact. "The regexes end at the opening paren" is the whole bug in one line, and noting that SCOPE.Component was never affected because contractRE runs to the { is the kind of detail that saves the next person from widening the fix past where it belongs.

declSignature is the right shape: one pass, paren depth, stop at the first { or ; at depth zero. That picks up the parameter list, the mutability/visibility keywords and the returns (...) clause without special-casing any of them. And the forEach(function(uint256) external returns (bool) cb) test is exactly the case I'd have wanted to see — a function-type parameter is precisely where naive paren matching falls over, and you covered it without being asked.

Two judgement calls I particularly appreciated:

You documented the limitation instead of hiding it. onlyRole("ADMIN)ROLE") coming out as onlyRole( ) is a real consequence of scanning scrubbed source, and you wrote it down and pinned it with a test. A known, tested limitation is worth far more than an undocumented edge case someone trips over in six months.

The fallback degrades instead of running away. A declaration with no terminator returning the previous span means malformed source gets the old behaviour rather than swallowing the rest of the contract. That's the conservative choice and it's the right one.

The before/after across seven production repositories — 5255 truncated to 0, entity ids unchanged, no field other than signature differing — is more evidence than most changes this size arrive with. Thank you for measuring it rather than asserting it.

And thank you for updating docs/coverage/registry.json and the capability map, including keeping the registry cell's original "issue": "5371" and only touching notes and verified_at. That step gets missed often, and getting it exactly right without prompting is appreciated.

Merging as-is. Nothing needed from you.