perf: speed up action startup by dropping per-call uv overhead

[vanandrew]

Summary

• Swap install from uv venv + uv pip install to uv tool install, putting difftrace on PATH and removing uv run wrapper overhead on every subsequent invocation.
• Collapse the five uv run python3 output-parsing calls (1 json.tool pretty-print + 4 separate -c parses for affected / matrix / test_all / count) into one python3 heredoc that emits all values together. Uses a tmpfile + four reads — portable across bash 3.2 and 5.x.
• Net effect: ~1s of venv-activation overhead shaved off each action run; install step is also slightly simpler.

Test plan

• Test Action workflow (.github/workflows/test-action.yml) passes for all matrix legs: single-lock, multi-lock, multi-lock downstream fan-out, single-lock downstream fan-out, inputs pass-through, and the missing-lock negative case.
• Output JSON shapes unchanged (affected list, matrix with either {"package": [...]} or {"include": [...]}, has_affected, test_all).
• Step summary table still renders correctly.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.08%. Comparing base (01c0d3c) to head (de8fdc2).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main       #9   +/-   ##
=======================================
  Coverage   98.08%   98.08%           
=======================================
  Files           6        6           
  Lines         470      470           
=======================================
  Hits          461      461           
  Misses          9        9           

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