Stop Windows test runs from running out of loopback ports (#1839)

Windows has no native socketpair, so the self-pipe of every asyncio event
loop is a loopback TCP connection, and closing it leaves a port in
TIME_WAIT for two minutes. With a new event loop for each test, a run
that gets through about 16,000 tests within two minutes uses up the
dynamic port range, and the Windows test jobs hang until the 6-hour job
limit.

This PR runs the tests on one event loop per session, so a full run
opens about 100 socket pairs instead of about 26,000, and runs the
Windows jobs with xdist again.

In details:
- `asyncio_default_fixture_loop_scope` and
  `asyncio_default_test_loop_scope` are `session`.
- CI: Windows runs `pytest -n auto` like the other OSes, and the test
  jobs get a 30-minute limit.

Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -54,6 +54,7 @@ jobs:
     name: Python ${{ matrix.python-version}} on ${{ matrix.os }}${{ fromJSON('[" (extras)", ""]')[matrix.extras == ''] }}
     needs: lint
     runs-on: ${{ matrix.os }}
+    timeout-minutes: 30
     strategy:
       matrix:
         python-version: [3.11, 3.12, 3.13, 3.14, 3.15]
@@ -80,11 +81,7 @@ jobs:
         id: run-pytests-with-code-coverage
         shell: bash
         run: |
-          if [[ "${{ runner.os }}" == "Windows" ]]; then
-            uv run pytest -n0 --cov kasa --cov-report xml
-          else
-            uv run pytest -n auto --cov kasa --cov-report xml
-          fi
+          uv run pytest -n auto --cov kasa --cov-report xml
 
       - name: Upload Code Coverage to Codecov
         id: upload-code-coverage-to-codecov