Try a way to workaround windows CI failues

... which hopefully works better

.github/workflows/ci.yml

@@ -42,17 +42,19 @@ jobs:
         key: ${{ runner.os }}-${{ matrix.ghc }}
     # We rebuild tests several times to avoid intermittent failures on Windows
     # https://github.com/haskell/actions/issues/36
+    # We also use --enable-tests and --enable-benchmarks to avoid
+    # test and bench commands from reconfiguring and thus rebuilding.
     - name: Test
       run: |
         cabal sdist -z -o .
         cabal get bytestring-*.tar.gz
         cd bytestring-*/
-        bld() { cabal build bytestring:tests; }
+        bld() { cabal build bytestring:tests --enable-tests --enable-benchmarks; }
         bld || bld || bld
-        cabal test --test-show-details=direct all
+        cabal test --enable-tests --enable-benchmarks --test-show-details=direct all -j1
     - name: Bench
       run: |
         cd bytestring-*/
-        cabal bench --benchmark-option=-l all
+        cabal bench --enable-tests --enable-benchmarks --benchmark-option=-l all
     - name: Haddock
       run: cabal haddock