Summary

Fixes #14695.

When a quoted annotation like "'in\x74'" is parsed as a complex annotation (because the raw content differs from the parsed content due to escape sequences), parse_complex_type_annotation produces an inner Expr::StringLiteral("int") and relocate_expr gives it the same text range as the outer annotation.

The ParsedAnnotationsCache is keyed by annotation.start(), so when TypingTarget::try_from_expr later encounters this inner StringLiteral and calls parse_type_annotation again, the cache returns the same result. This creates an infinite recursion path:

try_from_expr(StringLiteral) -> parse_type_annotation (cache hit) -> ForwardReference(StringLiteral) -> contains_any / contains_none -> try_from_expr(StringLiteral) -> ...

Fix

In try_from_expr, when a parsed forward reference resolves to another StringLiteral, return None instead of wrapping it as a ForwardReference. A string literal inside a string annotation (e.g. 'int' inside "'int'") is never a valid type -- it's a literal value, not a type name -- so there is no reason to recurse into it.

Test Plan

Added a regression test to annotation_presence.py with the exact reproducer from the issue (def f(x: "'in\x74'"): pass). The test previously caused a stack overflow and now correctly produces ANN401 + ANN201 diagnostics.

All existing flake8_annotations and implicit_optional tests continue to pass.