perf(api): trim Python overhead on the plain decode fast path

[ZhuchkaTriplesix]

Problem

For HS256 the Python wrapper costs about as much as the native work:

	µs/call
_oxyjwt.decode (native)	1.30
oxyjwt.decode (fast path)	2.14

The fast path in PyJWT.decode / decode_complete (python/oxyjwt/api_jwt.py:206-218, :257-272) calls _is_plain_decode (9 args), _require_detached_payload_for_rfc7797, and _validate_claims_default, which calls time.time() and re-checks exp even though Rust already validated it.

Proposal

• Inline _is_plain_decode and the ".." in token pre-check into the fast path.
• In _validate_claims_default, only re-check exp in Python when it is not an int (for ints Rust exp <= now_int and Python exp <= time.time() are equivalent; for floats Rust rounds while Python truncates, so that case must stay for PyJWT parity).
• Call time.time() only when iat or a non-int exp is present.

Acceptance criteria

• HS256 oxyjwt.decode overhead over native reduced by ≥ 30%
• tests/test_decode_fast_path.py, tests/test_parity_pyjwt.py green; add a test for float exp boundary parity

Files

• python/oxyjwt/api_jwt.py

Branch

issue/<N>-python-fast-path-overhead from dev

[ZhuchkaTriplesix]

Fixed in #128 (merged to dev).