Close LLM client httpx pool before asyncio.run exits

[daniel-thom]

Summary

• Add aclose() to the LLMClient protocol and implement it on AnthropicLLMClient and _OpenAICompatibleClient (delegates to the SDK's async close(), which drains the underlying httpx.AsyncClient connection pool).
• Wrap the body of cli.run_ask_pipeline, generate._run, and tidy._call in try/finally: await llm_client.aclose() so the pool is closed inside the same event loop that uses it — before asyncio.run shuts the loop down.

Without this, the SDK's httpx client gets garbage-collected after the loop is already closed, and its scheduled aclose() trips RuntimeError: Event loop is closed. asyncio prints that as a "Task exception was never retrieved" traceback. Particularly noisy with datasight ask --files, which spawns one event loop per question — every row in a benchmark batch produced the traceback.

The fix also covers datasight generate and datasight tidy review, which had the same teardown pattern.

Test plan

• pytest -m "not integration" — 1492 passed.
• prek run --all-files — ruff / ruff-format / ty all clean.
• New tests/test_llm.py cases assert that AnthropicLLMClient.aclose() and OllamaLLMClient.aclose() actually invoke the SDK's close().
• Manual: run datasight ask --files=questions.txt and confirm no "Event loop is closed" traceback at the end of the run.

🤖 Generated with Claude Code

[Copilot]

Pull request overview

This PR adds an explicit async teardown hook to the LLM client abstraction and ensures CLI code paths close the underlying SDK/httpx connection pool before asyncio.run() shuts down the event loop, preventing noisy RuntimeError: Event loop is closed “Task exception was never retrieved” tracebacks during garbage collection.

Changes:

• Extend the LLMClient protocol with aclose() and implement it for Anthropic and OpenAI-compatible backends.
• Wrap key CLI async entrypoints (ask, generate, tidy review) in try/finally to await llm_client.aclose() within the same event loop.
• Update/add tests and test stubs to satisfy the new protocol and assert the SDK close() method is awaited.

Reviewed changes

Copilot reviewed 11 out of 11 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
src/datasight/llm.py	Add LLMClient.aclose() and implement pool shutdown for Anthropic and OpenAI-compatible clients.
src/datasight/cli.py	Ensure run_ask_pipeline closes the LLM client before loop teardown via try/finally.
src/datasight/cli_commands/generate.py	Ensure the generate command closes the LLM client within the _run() event loop.
src/datasight/cli_commands/tidy.py	Ensure the tidy review LLM call closes the LLM client within the _call() event loop.
tests/test_llm.py	Add async tests asserting aclose() awaits the underlying SDK close().
tests/test_cli_tools.py	Add/update LLM stubs to provide aclose() for new teardown logic in pipelines.
tests/test_cli_commands.py	Update CLI test LLM stub to implement aclose().
tests/test_verify.py	Update fake LLM clients to implement aclose() for protocol compatibility.
tests/test_tidy_review.py	Update fake LLM client to implement aclose().
tests/test_generate_persist_db.py	Update stub LLM client to implement aclose().
tests/test_agent_extra.py	Update fake LLM client to implement aclose().

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[codecov]

Codecov Report

❌ Patch coverage is 76.52174% with 27 lines in your changes missing coverage. Please review.
✅ Project coverage is 86.86%. Comparing base (b635119) to head (7e432ae).

Files with missing lines	Patch %	Lines
src/datasight/cli_commands/generate.py	78.18%	12 Missing ⚠️
src/datasight/cli_commands/verify.py	0.00%	12 Missing ⚠️
src/datasight/cli.py	92.50%	3 Missing ⚠️

❌ Your patch check has failed because the patch coverage (76.52%) is below the target coverage (80.00%). You can increase the patch coverage or adjust the target coverage.

Additional details and impacted files

@@            Coverage Diff             @@
##             main      #70      +/-   ##
==========================================
- Coverage   86.87%   86.86%   -0.01%     
==========================================
  Files          61       61              
  Lines       11539    11552      +13     
==========================================
+ Hits        10024    10035      +11     
- Misses       1515     1517       +2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.