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

[ZhuchkaTriplesix]

Problem

json_to_bound (rust/src/claims.rs:122-126) allocates a fresh PyString for every dict key on every decode, including the same standard claim / header names each time.

Proposal

Use pyo3::intern! (or a small static table) for common keys — exp, iat, nbf, sub, aud, iss, jti, alg, typ, kid — and fall back to a new PyString otherwise. Interned keys also make subsequent dict.get in Python hit the pointer-equality fast path.

Follow-up to consider (measure first): deserialize payload JSON straight into Python objects via a serde visitor, skipping the intermediate serde_json::Value (claims validation would need to move to a borrowed view or a lightweight struct).

Acceptance criteria

• Common keys are interned
• Micro-benchmark (decode of a typical 6–10 claim payload) shows improvement; numbers in PR
• Tests green

Files

• rust/src/claims.rs

Branch

issue/<N>-intern-claim-keys from dev

[ZhuchkaTriplesix]

Fixed in #129 (merged to dev).