test(e2e): stop the wine teardown from failing a passing test

`cross-build-test` went red on main at 0006ce6 (run 31336526583) with a log
that ends:

    OK
    rm: cannot remove '/tmp/tmp.XgyezkVtrY/wineprefix': Directory not empty

The test passed. `wine` leaves a wineserver running after the process it
launched exits, and that server keeps writing into $WINEPREFIX, so the
EXIT trap's `rm -rf` races it. Under `set -e` the failing trap becomes the
script's exit status, so a green test reports as a red job — and the
verdict it reports is about a temp directory, not about mcpp.

Shut the server down and wait for it before deleting, and let cleanup
keep the status the test already earned. Cleanup runs after the verdict is
printed; it has no business changing it.

Author: speak-agent

tests/e2e/102_mingw_cross_wine.sh

@@ -9,7 +9,28 @@
 set -e
 
 TMP=$(mktemp -d)
-trap "rm -rf $TMP" EXIT
+
+# `wine` leaves a wineserver running after the process it launched exits, and
+# that server keeps writing into $WINEPREFIX. A plain `rm -rf` in the EXIT trap
+# therefore races it and can fail with "Directory not empty" — AFTER the test
+# has already printed OK. Under `set -e` that failing trap becomes the script's
+# exit status, so a passing test reports as a red job (observed on main at
+# 0006ce6, run 31336526583: the log ends `OK` then `rm: cannot remove
+# '…/wineprefix': Directory not empty`).
+#
+# So: shut the server down first and wait for it, and never let cleanup decide
+# the verdict. By the time this runs the verdict is already printed; a failure
+# to delete a temp directory is not a statement about mcpp.
+cleanup() {
+    local status=$?
+    if [[ -n "${WINEPREFIX:-}" ]] && command -v wineserver >/dev/null 2>&1; then
+        wineserver -k >/dev/null 2>&1 || true
+        wineserver -w >/dev/null 2>&1 || true
+    fi
+    rm -rf "$TMP" 2>/dev/null || true
+    return $status
+}
+trap cleanup EXIT
 cd "$TMP"
 
 "$MCPP" new crosswin