fix(compat): add usage retention pruning and Anthropic error responses

[mojomast]

Summary

• add simple usage retention controls via USAGE_RETENTION_DAYS with startup pruning and admin-triggered prune endpoint (POST /api/admin/usage/prune)
• return Anthropic-style JSON error payloads directly from /v1/messages and /v1/messages/count_tokens without FastAPI detail wrapping
• document retention and admin-only quota visibility in README and .env.example
• add tests for Anthropic error response shape, retention pruning, and admin actor guard regression

Verification

• PYTHONPATH=src .venv/bin/python -m py_compile src/proxy_app/usage_recorder.py src/proxy_app/anthropic_errors.py src/proxy_app/routers/admin_api.py src/proxy_app/main.py tests/test_api_compat.py tests/test_usage_retention.py tests/test_auth.py
• PYTHONPATH=src .venv/bin/pytest -q tests/test_api_compat.py tests/test_usage_retention.py tests/test_auth.py
• PYTHONPATH=src .venv/bin/pytest -q

---

Important

Adds usage retention controls and Anthropic-style error responses, with tests and documentation updates.

• Behavior:
  • Adds USAGE_RETENTION_DAYS for usage event retention, with startup pruning and POST /api/admin/usage/prune for manual pruning.
  • Updates /v1/messages and /v1/messages/count_tokens to return Anthropic-style JSON error payloads without FastAPI detail wrapping.
• Documentation:
  • Documents usage retention and admin-only quota visibility in README.md and .env.example.
• Testing:
  • Adds tests for Anthropic error response shape, retention pruning, and admin actor guard regression.

^{This description was created by for 68d0163. You can customize this summary. It will automatically update as commits are pushed.}

[mojomast]

Superseded by updates pushed to #123 to keep all MVP changes in one review thread.