Restore lexer depth on the unclosed-bracket error path

[greymoth-jp]

Parsing a format description with an unclosed nested bracket ([a[, [optional [, [hour[) panics in debug builds via debug_assert!(self.context().is_component()), reached through consume_whitespace.

The opening bracket increments self.depth. On the branch where no closing bracket is found, the lexer returns UnclosedOpeningBracket without restoring self.depth. consume_component discards that error and re-derives the unclosed-bracket error itself, so the lexer keeps going with depth one too high; context() then reports the wrong context and the assert fails.

The fix decrements self.depth before returning, matching the entry state like the other early returns in this function. Release builds skip the assert, so the observable symptom is a debug-build panic; the underlying issue is the leftover depth on this error path.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 87.4%. Comparing base (8faa889) to head (4712e1e).
⚠️ Report is 5 commits behind head on main.

Additional details and impacted files

@@          Coverage Diff          @@
##            main    #791   +/-   ##
=====================================
  Coverage   87.4%   87.4%           
=====================================
  Files        104     104           
  Lines      13982   13983    +1     
=====================================
+ Hits       12223   12224    +1     
  Misses      1759    1759           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jhpratt]

Please rewrite the description. LLM-generated writing is not acceptable, and frankly I have no idea what it even means. There is no such thing as a "native" parser. Let me know when the description is rewritten and I will take a look.

[greymoth-jp]

Rewrote the description and title. You're right that "native parser" was meaningless; it's a debug_assert in the format-description lexer. The opening bracket increments depth and the unclosed-bracket error path didn't restore it, so a later context() assert trips on inputs like [a[. Happy to add a regression test covering those inputs if you'd want it in the same PR.

[jhpratt]

Yes, please add a regression test for the inputs. They can go in the parse_format_description.rs test file. Also, if you could ensure that an LLM is not listed as co-author in the commit, that would be preferred.

[greymoth-jp]

Added regression cases for [a[ and [hour[ to nested_v2_error_unclosed. Without the depth restore they trip the debug_assert!(self.context().is_component()) in consume_whitespace; with it they report UnclosedOpeningBracket at index 0 like the existing cases. [optional [ was already covered there and, owing to its trailing space, skips that consume_whitespace call, so it never reached the assert. The commit no longer lists a co-author.

[jhpratt]

Looks good; thanks!