gh-155962: Specialize LOAD_ATTR after replacing an instance's __dict__

[koxudaxi]

Replacing an instance's __dict__ prevents LOAD_ATTR from specializing on Python 3.14 and main. The same case specializes to LOAD_ATTR_WITH_HINT on Python 3.13.

instance_has_key() used shared keys whenever the type supported inline values, even after replacing an instance's __dict__ had invalidated its inline values. Check the per-instance state and use the managed dictionary instead. General-key dictionaries remain unsupported by the hint opcodes.

This affects Pydantic v2 because pydantic-core installs validated fields by replacing the model's __dict__.

Benchmark

Fresh-process datamodel-code-generator CLI generation using a 500-definition JSON Schema, Pydantic v2 output, the built-in formatter, and a temporary output file:

	Median
main	354.5 ms
this PR	327.7 ms

This is a 26.8 ms (7.6%) reduction. The results are medians from 11 alternating-order samples on macOS arm64.

Tests

Tests cover replacement dictionaries with stale shared keys, unsupported general-key dictionaries, and concurrent dictionary replacement.

Fixes #155962

• Issue: LOAD_ATTR does not specialize after replacing an instance's __dict__ #155962