Share Responses body size limits

[franciscojavierarceo]

Part of #213.

Summary

• add a shared Responses body-size alias for filters that inspect or transform Responses request bodies
• switch Responses validation to the shared 64 MiB Responses limit
• keep the format classifier on the generic JSON inspection default

Stack

• base of Add Responses request translation core #702
• followed by Add Responses translation semantics #703, Add Responses to Chat Completions filter #696, and Add Responses SSE framing coverage #677

Validation

• cargo test -p praxis-proxy-filter --lib --features ai-inference openai::responses
• cargo test -p praxis-tests-integration openai_responses_to_chat_completions
• make lint

[praxis-bot]

Share Responses body size limits

Clean constant-extraction refactor. The validate filter's local 64 MiB constant moves to the shared body/limits.rs module as OPENAI_RESPONSES_BODY_MAX_BYTES, and the format classifier's local 10 MiB constant is replaced by the existing shared DEFAULT_JSON_BODY_MAX_BYTES. No behavioral changes.

Severity	Count
Medium	1

The OPENAI_RESPONSES_BODY_MAX_BYTES re-export in body/mod.rs bypasses limits -- consumers can import from body::OPENAI_RESPONSES_BODY_MAX_BYTES instead of body::limits::OPENAI_RESPONSES_BODY_MAX_BYTES. The store filter already imports MAX_JSON_BODY_BYTES directly from body::limits:: (line 60 of store/filter.rs), so the two access patterns are inconsistent across the codebase. Not blocking, but worth aligning in a follow-up.