fix(parsing): drop TextFormatT parameterization in parse_response to fix memory leak (#3084)

[MukundaKatta]

Why

Closes #3084.

AsyncResponses.parse() leaks pydantic schema objects without bound. The issue author captured it with a flame graph — every call allocates a fresh Rust-backed SchemaValidator/SchemaSerializer that never gets freed.

Root cause

parse_response() called construct_type_unchecked with three types parameterized by a free module-level TypeVar:

construct_type_unchecked(type_=ParsedResponseOutputText[TextFormatT], ...)
construct_type_unchecked(type_=ParsedResponseOutputMessage[TextFormatT], ...)
construct_type_unchecked(type_=ParsedResponse[TextFormatT], ...)

Pydantic can't resolve a free TypeVar, so model_rebuild(raise_errors=False) returns False. MockCoreSchema._built_memo only caches the rebuilt schema when model_rebuild succeeds — so the cache is never populated and a new schema is allocated on every call.

Fix

Drop the [TextFormatT] parameterization at the type_= argument. At runtime Python's generics are erased anyway — the constructed object's type is identical either way. Only the pydantic schema rebuild path differs:

	Before	After
type_ arg	ParsedResponse[TextFormatT]	ParsedResponse
model_rebuild	always returns False	succeeds
_built_memo	never populated	populated once, reused
Per-call allocation	heavy Rust objects	cache hit

This is exactly what the issue author diagnosed.

Verification

• Ran the existing tests/lib/responses/test_responses.py suite locally: 5/5 passing — no functional regression.
• The change is purely about schema caching behavior. Callers see no difference in returned objects or their types.

Scope

Minimal — 3 sites in one file, inline comments pointing back to #3084 for maintainability. No public API change, no test changes needed (existing tests cover the parse path).