🐛 Fix unhashable `Annotated` type in `get_definitions` OpenAPI schema generation

`get_definitions()` uses a set comprehension to collect field annotations
for deduplication. However, on Python 3.14 with Pydantic v2.13+, some
`Annotated` types contain `FieldInfoMetadata` objects that are not
hashable, causing a `TypeError`:

    TypeError: cannot use 'typing._AnnotatedAlias' as a set element
    (unhashable type: 'FieldInfoMetadata')

This is triggered when routes use `Annotated` parameters with metadata
such as `Query(max_length=..., description=...)` or similar constructs
that produce unhashable Pydantic field metadata.

Fix: use `id()` for identity-based deduplication instead of relying on
`__hash__` of annotation objects. Since annotations are interned type
objects, `id()` comparison is semantically equivalent for this filtering
purpose.

Co-Authored-By: Claude Opus 4.6 (1M context) <[email protected]>

fastapi/_compat/v2.py

@@ -317,9 +317,9 @@ def get_definitions(
         for model in flat_serialization_models
     ]
     flat_model_fields = flat_validation_model_fields + flat_serialization_model_fields
-    input_types = {f.field_info.annotation for f in fields}
+    input_type_ids = {id(f.field_info.annotation) for f in fields}
     unique_flat_model_fields = {
-        f for f in flat_model_fields if f.field_info.annotation not in input_types
+        f for f in flat_model_fields if id(f.field_info.annotation) not in input_type_ids
     }
     inputs = [
         (