emrg: add Windows pytest matrix to test.yml (v0.2.29 CI gap)

[argszero]

Summary

Close the CI coverage gap behind the two v0.2.29 Build Release failures (31604108964 / 31607172268):

• Before: Windows pytest only ran in the tag-triggered build-release.yml (test.yml was ubuntu-only). Windows-only test regressions — e.g. the FakeGitRun case-sensitive git.exe/git.EXE bug (emrg: fix v0.2.29 Build Release gates — Windows git.EXE fake + filter-aware icon decode #723) that failed 8 scheduler tests — were only caught at release time, after the tag was pushed.
• After: new test-windows job runs the full pytest suite on windows-2025 for every push/PR, mirroring the exact invocation already proven green in Build Release run 31608141280 (setup-uv → uv sync → uv run pytest tests/ -v).

Changes

• .github/workflows/test.yml: add test-windows job (checkout → setup-uv python 3.13 → uv sync → pytest)
• Agent.md: CI section notes the ubuntu + windows-2025 pytest matrix

No test count changes (workflow-only + doc line).

[argszero]

✅ LGTM — cycle

The v0.2.29 release failures (31604108964 / 31607172268) proved Windows-only test regressions — like the FakeGitRun git.EXE case-sensitivity bug (#723) — only surfaced at tag-build time. Moving Windows pytest into the per-PR matrix closes that gap with the exact invocation already proven green in Build Release 31608141280. actionlint gate passed in CI (31609202755).

Non-blocking nit: the Agent.md CI line cites "#724" as the lesson reference — should be this PR (#725). Cosmetic doc footnote; no functional impact.

[argszero]

✅ LGTM — cycle (CI: test + test-windows both pass; actionlint gate green on the workflow change)

[argszero]

✅ LGTM — cycle (3rd consecutive ✅)

Windows pytest matrix in test.yml closes the v0.2.29 CI gap (Windows-only regressions like git.EXE #723 now fail at PR time). Head 006fd61 unchanged; test + test-windows both SUCCESS in CI; actionlint gate green. Non-blocking nit remains: Agent.md lesson ref says #724, should be #725 (cosmetic doc footnote). Merge condition met (3 consecutive ✅ from different cycles, no ❌ in between).