fix(reflect): cap structured output retry budget

[koriyoshi2041]

Summary

• cap the second-pass reflect_structured extraction retry budget so provider-specific empty structured-output responses do not consume the full reflect timeout
• keep the existing graceful degradation behavior: the natural-language reflect answer still returns when structured extraction fails
• add a focused regression test for the short retry budget

Context

This is motivated by #2431, where plain reflect can succeed while response_schema triggers repeated empty provider responses with finish_reason=length.

Tests

• uv run pytest tests/test_reflect_agent.py::TestReflectStructuredOutput::test_structured_output_uses_short_retry_budget -q
• uv run ruff check hindsight_api/engine/reflect/agent.py tests/test_reflect_agent.py
• uv run ruff format --check hindsight_api/engine/reflect/agent.py tests/test_reflect_agent.py
• git diff --check

Note: full uv run pytest tests/test_reflect_agent.py -q reached 50 passing tests, then hit local environment setup errors unrelated to this patch: missing HINDSIGHT_API_LLM_API_KEY for real-LLM tests and a local SOCKS proxy requiring socksio during Hugging Face model initialization.