[ty] Improve diagnostics for syntax errors in forward annotations

[AlexWaygood]

Summary

Fixes astral-sh/ty#1627.

Here's an example diagnostic with the current release of ty:

On this branch, this diagnostic becomes:

The exact span of the node that creates the syntax error is now retained and highlighted in the diagnostic.

Implementation

Propagating the range of the node inside the string annotation into the diagnostic is trivial. However, naively implementing that quickly revealed that this would make diagnostics like this unsuppressable:

x: """list[
    yield from range(42)
]"""

The primary range of the diagnostic is now specifically the yield from range(42) part of the string rather than the string node as a whole. But I cannot add a ty: ignore comment that is either on or above the yield from range(42) part of the string -- the "comment" there would just become part of the string. In order to ensure that these diagnostics are suppressible (and testable in mdtest), therefore, it was necessary to add a way to attach a custom primary range to a diagnostic separate to that diagnostic's suppression range. This is something we've talked about doing lots of times before, so I hope it isn't too controversial.

This in itself was also fairly trivial, but I soon realised that in order for mdtest to be able to recognise # error assertions correctly, we would need to retain the custom suppression ranges for diagnostics iff the testing feature was activated. This ended up feeling slightly icky (lots of #[cfg(feature = "testing")] attributes everywhere), but I'm not sure I see another way.

The second commit of this PR improves the consistency of our parser error messages in general when it comes to capitalization.

Test Plan

Mdtests extended and updated

[astral-sh-bot]

Typing conformance results

No changes detected ✅

Current numbers

The percentage of diagnostics emitted that were expected errors held steady at 89.36%. The percentage of expected errors that received a diagnostic held steady at 85.49%. The number of fully passing files held steady at 88/134.

[astral-sh-bot]

Memory usage report

Memory usage unchanged ✅

[astral-sh-bot]

ecosystem-analyzer results

No diagnostic changes detected ✅

Full report with detailed diff (timing results)

[astral-sh-bot]

ruff-ecosystem results

Linter (stable)

✅ ecosystem check detected no linter changes.

Linter (preview)

✅ ecosystem check detected no linter changes.

Formatter (stable)

✅ ecosystem check detected no format changes.

Formatter (preview)

✅ ecosystem check detected no format changes.

[AlexWaygood]

this was added automatically by uv run crates/ty_python_semantic/mdtest.py. I assume it's a change in newer versions of uv?

[MichaReiser]

Does is_range_reachable work with ranges that point into string annotations?

[AlexWaygood]

I don't see why it wouldn't: the range inside a string annotation is still relative to the start of the file and is_range_reachable just uses contains_range:

ruff/crates/ty_python_semantic/src/reachability.rs

Line 992 in c12e77f

entry_range.contains_range(range) && !is_reachable(db, use_def, constraint)

[AlexWaygood]

Okay, I've removed mdtest's custom assertion-matching mechanism. The relevant APIs in the mdtest crate now accept closures that determine whether an assertion should match a given diagnostic, which allows us to hook into ty's error-suppression logic rather than handrolling something totally different in mdtest. I think this should hopefully be extensible for the ongoing project to use mdtests in Ruff too