perf(claims): intern common claim/header keys in json_to_bound (#123)

[ZhuchkaTriplesix]

Problem

json_to_bound (used by every decode/decode_complete call) allocated a fresh PyString for every dict key on every call, including the same standard claim/header names every time.

Fix

The ten standard names — exp, iat, nbf, sub, aud, iss, jti, alg, typ, kid — now come from pyo3::intern!, which caches each key in a call-site-local static instead of allocating it fresh. Interning also registers the string in CPython's own intern table, so a later payload.get("exp") on the Python side (a str literal, which CPython auto-interns) can hit the identity-comparison fast path during dict lookup instead of a full string comparison. Any other key still falls back to an ordinary, uninterned PyString, unchanged from before.

No behavioral change.

Results

Native decode, 8-claim payload (sub, iat, exp, nbf, iss, jti, plus two custom keys), release build, averaged over repeated runs:

	before	after
native decode	~2.05 µs	~1.92 µs (~6% less)

Modest, as expected — allocation is a small fraction of total decode cost next to base64/JSON/signature verification — but real and reproducible.

Tests

Added test_standard_claim_key_is_interned, test_standard_header_key_is_interned and test_custom_claim_key_is_not_interned (tests/test_encode_decode.py): each standard name must decode back as the exact same interned object (is, not just ==), while a key outside the list must still decode correctly as an ordinary, uninterned string.

(Considered a #[cfg(test)] Rust unit test instead/also, but this crate's existing Rust tests are pure-logic and never touch the Python API — adding one that does would need the auto-initialize pyo3 feature as a new dev-dependency, which changes cargo test's requirements project-wide for a change this contained. The Python-level identity test gives the same guarantee through the existing test infrastructure.)

Checklist

• Common keys are interned
• Micro-benchmark (8-claim payload decode) shows improvement; numbers above
• cargo build/clippy --all-targets -D warnings/fmt --check/cargo test clean with both aws_lc_rs (default) and --no-default-features --features rust_crypto
• Full pytest (270 passed, 1 skipped) and mypy clean
• No behavioral change

Closes #123.