cache gateway lookup and async checkpoint I/O

[cchinchilla-dev]

Summary

• Cache resolved provider candidates in ProviderGateway._get_candidates() to avoid rebuilding the fallback list on every complete()/stream() call; cache invalidates on register()
• Wrap save_checkpoint and from_checkpoint blocking I/O with anyio.to_thread.run_sync so disk operations don't block the event loop during concurrent step execution

Test plan

• uv run pytest — 690 pass, 7 e2e deselected
• uv run mypy src/ — no issues
• uv run ruff check src/ tests/ — all checks passed

Closes #9
Closes #14

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[Copilot]

Pull request overview

This PR improves runtime performance and async responsiveness by (1) caching provider candidate resolution in ProviderGateway and (2) moving checkpoint file I/O off the event loop via anyio.to_thread.run_sync.

Changes:

• Cache _get_candidates() results per model in ProviderGateway, invalidating the cache on register().
• Wrap checkpoint read/write operations in StateManager.save_checkpoint() / StateManager.from_checkpoint() with anyio.to_thread.run_sync.
• Update state checkpoint tests to await the now-async StateManager.from_checkpoint().

Reviewed changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
src/agentloom/providers/gateway.py	Adds per-model candidate caching to avoid rebuilding fallback/provider lists on every request.
src/agentloom/core/state.py	Offloads checkpoint JSON read/write to a worker thread; makes from_checkpoint async.
tests/core/test_state.py	Updates checkpoint restore tests to await from_checkpoint.