fix(lint): preserve underscores and honor mixed_case_exceptions in pascal-case-struct

[gomesalexandre]

What

pascal-case-struct had two bugs, both in crates/lint/src/sol/naming.rs / crates/lint/src/sol/info/pascal_case.rs:

1. Underscores weren't preserved. check_pascal_case was the only one of the three naming helpers (check_pascal_case, check_screaming_snake_case, check_mixed_case) that didn't call the shared preserve_underscores helper. A struct named _PascalCase got flagged with a MachineApplicable suggestion to rename to PascalCase — colliding with an already-existing struct of that exact name in the same contract. The existing StructPascalCase.sol fixture was pinning exactly this broken suggestion (struct PascalCase at one line, struct _PascalCase flagged suggesting PascalCase at another).

2. mixed_case_exceptions was never consulted. mixed-case-function/mixed-case-variable both respect the configurable acronym allowlist (ERC, URI, etc. by default) via LintSpecificConfig, but pascal-case-struct never received the config at all — so struct ERC20Data gets flagged and suggested as Erc20Data.

Fix

• check_pascal_case now calls preserve_underscores, matching its two siblings.
• pascal_case.rs gained a PascalCaseStructPass wrapping Arc<LintSpecificConfig>, wired through register_lints!'s constructor mechanism exactly the way mixed_case.rs's two passes already do — including the same pre/post AsUpperCamelCase pattern-exception matching (adjusted for full PascalCase rather than lowerCamelCase).
• Updated StructPascalCase.sol/.stderr (stderr regenerated via cargo test -p forge --test ui -- --bless) to cover both fixes: an _pascalCase case (proves the underscore is preserved in a real suggestion, not just a no-op passthrough) and an ERC20Data case (proves the exception now applies). The previously-colliding _PascalCase case is no longer flagged at all, since it already satisfies PascalCase-with-preserved-underscore.

Testing

• cargo test -p forge-lint --lib — 11/11 passing, including new pascal_case_preserves_underscores unit tests.
• Full fixture UI suite (cargo test -p forge --test ui) — 97/97 passing after the bless regeneration; confirmed the blessed diff only touches the one line that changed (the other already-flagged invalid cases — pascalCase, pascalcase, pascal_case, PASCAL_CASE, PASCALCASE — are unaffected).
• Manually re-verified: s.find(pattern) on an empty pre/post, a pattern at the very start/end of the string, and a pattern longer than the input all resolve safely (no panic, no false match) — same as the pre-existing mixed_case.rs implementation this mirrors.

Note on review

Ran a synchronous Codex adversarial pass on this diff. It got partway through independently cross-referencing the exact mixed_case.rs pattern this PR mirrors before hitting a transient "model at capacity" error and never returned a final verdict. Did a manual self-review of the edge cases it was mid-way through checking (pattern-boundary matching, empty pre/post) rather than re-run it — noting this honestly rather than claiming a completed automated review.

closes nothing — filed independently, dupe-checked clean (no existing issue/PR on this).

[github-actions]

✅ Changelog found

The deterministic check will validate the changed entry.