fix(anthropic_response): fail translation on malformed tool arguments (#550)

[leseb]

Summary

When translating an OpenAI Chat Completions response into Anthropic Messages format, tool-call arguments that were not a valid JSON-encoded object were silently replaced with an empty object ({}), which could make an Anthropic client execute a tool with missing or default parameters. Translation now fails on malformed/non-object arguments instead of fabricating an executable tool_use block, and because the upstream status headers are already committed by the time the body is parsed, the original upstream response is preserved unchanged rather than emitting a misleading 200 OK error. This is the smallest complete change: a Result-returning transform plus unit, functional proxy, and controlled synthetic inference-fixture coverage.

Related issue

Closes #550

Validation

• make lint — clippy, rustfmt, dependency, docs, example, and check-inference gates all pass.

• make build

• cargo test -p praxis-ai-apis --lib -- anthropic::to_openai::response — 21 passed, incl. invalid_tool_call_arguments_fail_transformation and non_object_tool_call_arguments_fail_transformation.

• make test-inference-fixtures — coverage tests + all_inference_fixtures_replay pass.

• cargo test -p praxis-tests-integration --test suite anthropic_to_openai_preserves_response_with_malformed_tool_arguments

• Unit tests

• Integration or functional tests

• make lint

Checklist

• I reviewed every changed line and can explain the change.
• New capabilities include an example config and functional example test. (N/A — bug fix; reuses the existing anthropic/messages-to-openai.yaml example with a new functional test.)
• User-facing behavior and generated documentation are updated.
• Performance-sensitive changes include appropriate benchmark or load-test evidence. (N/A)
• Commits are signed and include a Signed-off-by trailer.

Breaking changes

None. Previously a malformed tool-argument response produced a fabricated empty-input tool_use block; it now preserves the original upstream response so no valid tool_use is invented. Valid responses are unaffected.