ci(tests): split windows unit d-p shard to cut wall-clock

Previous run showed `unit d-p` as the slowest shard at ~2m01s.
Split into `unit d-g` (6 files, ~820 LOC) and `unit h-p` (4 files,
~848 LOC, includes the heavy `helpers_test.sh` and `parallel_test.sh`).
Wall-clock target drops to ~1m per shard. Adds one runner, halves
Windows completion time.

Author: Chemaclass

.github/workflows/tests.yml

@@ -119,8 +119,10 @@ jobs:
             test_path: "tests/unit/[a-b]*_test.sh"
           - name: "unit c + coverage"
             test_path: "tests/unit/ch*_test.sh tests/unit/cl*_test.sh tests/unit/console*_test.sh tests/unit/cu*_test.sh tests/unit/coverage_*_test.sh"
-          - name: "unit d-p"
-            test_path: "tests/unit/[d-p]*_test.sh"
+          - name: "unit d-g"
+            test_path: "tests/unit/[d-g]*_test.sh"
+          - name: "unit h-p"
+            test_path: "tests/unit/[h-p]*_test.sh"
           - name: "unit r-z"
             test_path: "tests/unit/[r-z]*_test.sh"
     steps: