[cdbartholomew]

Routine dependency maintenance for the root uv.lock.

Closes Dependabot alert: #2346

• setuptools 84.0.0 -> 84.2.1 (transitive, runtime dependency of torch; affected range >= 38.2.2, < 84.1.0). Lock regenerated with uv lock -P setuptools (uv 0.13.5); diff is the single package entry.

Local verification

• uv lock --check OK
• hindsight-api-slim: uv build OK, uv sync --frozen OK (Python 3.11), import setuptools, torch OK
• pytest tests/test_mcp_*.py — 488 passed, 4 skipped
• Full suite runs in test-api (root uv.lock is in the core path filter).

[cdbartholomew]

Merging over a known, unrelated CI failure. Evidence:

• Failing jobs here (run 40202431933, already re-run once, attempt 2): LLM acceptance (openai/gpt-4.1-nano) and LLM acceptance (litellmrouter/gpt-4.1-nano).
• Error signature: openai.NotFoundError: Error code: 404 ... The model gpt-4.1-nano has been deprecated (and the LiteLLM equivalent litellm.NotFoundError: OpenAIException - The model gpt-4.1-nano has been deprecated).
• The identical signature fails the same two jobs on PRs that do not touch uv.lock:
  • fix(worker): shutdown no longer waits out a long maintenance job #6092 (run 40204727062, jobs 121203193912 / 121203193880)
  • fix(retain): don't checkpoint documents a buffering retain session hasn't committed (#6099) #6100 (run 40301034674, jobs 121498789789 / 121498789709)
• main has since replaced the retired model in the acceptance matrix (ci: swap retired gpt-4.1-nano for gpt-6-luna in LLM acceptance matrix #6102, f29ba70).

The jobs that exercise this lockfile change all passed: test-api (1/3, 2/3, 3/3), build-api-python-versions (3.11–3.14), typecheck-api, Docker builds, test-pip-slim, test-embed.