fix(responses): disable parallel tools on first inference

[skamenan7]

Summary

Forces the agentic loop's first provider request to send parallel_tool_calls:false when the client omits the field or explicitly sends true. Provider-visible request mutations now mark the Responses state for outbound serialization, while an already-false request keeps byte-exact passthrough.

• agentic_loop marks changed provider-visible fields as requiring a request-body rebuild.
• openai_responses_proxy rebuilds only when the explicit request-body state requires it.
• Unit and configured integration tests cover omitted, explicit true, and byte-exact explicit false behavior.

Related issue

Closes #677

Validation

• Unit tests
• Integration or functional tests
• make lint

Validation run:

• cargo test -p praxis-ai-apis agentic_loop (46 passed)
• cargo test -p praxis-tests-integration --test suite examples::agentic_loop (6 passed)
• make test
• Built and launched the real praxis-ai server with the agentic-loop example configuration and a controlled local inference backend. The captured provider-bound requests show omitted and explicit true reaching first inference as false; explicit false retained the same request-body length and SHA-256 across the proxy.

Checklist

• I reviewed every changed line and can explain the change.
• New capabilities include an example config and functional example test.
• User-facing behavior and generated documentation are updated.
• Performance-sensitive changes include appropriate benchmark or load-test evidence.
• Commits are signed and include a Signed-off-by trailer.

Breaking changes

None.