fix(ci): kill stale platform-server before binding port

Cancelling or timing out a workflow run leaves the platform-server
process alive — the "Stop platform" step (line 335) is skipped.
On the next run, the OS may hand out :8080 (the default port) via
socket.bind(("", 0)), and platform-server then fails with:

    listen tcp :8080: bind: address already in use

Fix: add a pre-start step that scans /proc for any zombie
platform-server processes and kills them before the port probe or
start. Brief 1s sleep lets the kernel release sockets.

Uses only grep + kill — available on any Ubuntu runner without
extra tools. Safe: only targets platform-server binaries, ignores
other Go processes.

Refs: internal#374, issue #1046

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

.gitea/workflows/e2e-api.yml

@@ -69,6 +69,13 @@ name: E2E API Smoke Test
 # 2318) shows Postgres ready in 3s, Redis in 1s, Platform in 1s when
 # they DO come up. Timeouts are not the bottleneck; not bumped.
 #
+# Item #1046 (fixed 2026-05-14): Stale platform-server from cancelled runs
+#   lingers on :8080 after "Stop platform" step is skipped (workflow cancelled
+#   before reaching line 335). Added a pre-start "Kill stale platform-server"
+#   step (line 286) that scans /proc for zombie platform-server processes
+#   and kills them before the port probe or bind. Makes the ephemeral port
+#   probe + start sequence deterministic.
+#
 # Item explicitly NOT fixed here: failing test `Status back online`
 # fails because the platform's langgraph workspace template image
 # (ghcr.io/molecule-ai/workspace-template-langgraph:latest) returns
@@ -283,6 +290,33 @@ jobs:
           echo "PORT=${PLATFORM_PORT}" >> "$GITHUB_ENV"
           echo "BASE=http://127.0.0.1:${PLATFORM_PORT}" >> "$GITHUB_ENV"
           echo "Platform host port: ${PLATFORM_PORT}"
+      - name: Kill stale platform-server before start (issue #1046)
+        if: needs.detect-changes.outputs.api == 'true'
+        run: |
+          # Concurrent runs on the same host-network act_runner can leave a
+          # zombie platform-server from a cancelled/timeout run. Cancelled
+          # runs never reach the "Stop platform" step (line 335), so the
+          # old process lingers. Before picking a port or starting, scan
+          # /proc for any platform-server binary and kill it.
+          #
+          # "Pick platform port" uses socket.bind(("", 0)) which SHOULD get
+          # a free port from the OS, but the OS may hand out a port still
+          # in TIME_WAIT from a zombie. Killing zombie processes first makes
+          # both the port probe AND the bind deterministic.
+          #
+          # Safe: only kills platform-server; ignores other Go binaries.
+          # Uses only shell builtins + grep + kill — available on any Ubuntu
+          # runner without extra tools.
+          for pid in $(grep -l 'platform-server' /proc/*/comm 2>/dev/null | cut -d/ -f3); do
+            cmdline=$(cat "/proc/${pid}/cmdline" 2>/dev/null | tr '\0' ' ')
+            if echo "$cmdline" | grep -q 'platform-server'; then
+              echo "Killing stale platform-server pid ${pid}: ${cmdline}"
+              kill "$pid" 2>/dev/null || true
+            fi
+          done
+          # Brief pause to let the kernel release sockets
+          sleep 1
+          echo "Stale platform-server cleanup complete."
       - name: Start platform (background)
         if: needs.detect-changes.outputs.api == 'true'
         working-directory: workspace-server