ci(#869): raise full-test Windows lane timeout 15m->20m (#871)

The 'full test (windows-latest, 22)' job runs all 656 unit files in one lane and
has crept to ~14m+ over recent PRs (12m31s -> 13m45s -> 14m18s), so the 15m cap
started cancelling jobs mid-run and red-blocking product PRs even when every test
passes (the job is killed on wall-clock, not a test failure). Raise the full-test
job timeout to 20m to restore headroom. Durable fix (shard the Windows full unit
lane) tracked as a separate follow-up.

Closes #869

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

.github/workflows/test.yml

@@ -265,7 +265,10 @@ jobs:
     defaults:
       run:
         shell: ${{ matrix.shell }}
-    timeout-minutes: 15
+    # 20m, not 15m: the Windows full lane runs all 656 unit files in one job and
+    # has crept to ~14m+, so 15m started cancelling jobs mid-run (#869). Durable
+    # fix is to shard this lane — tracked separately.
+    timeout-minutes: 20
     env:
       GSD_PLUGIN_ROOT: .ci-gsd-plugin-root-disabled
     strategy: