fix(mcp): cancel the in-flight index before a force re-index instead of orphaning it

[EricSeastrand]

Problem

handleIndexCodebase with force: true, on a codebase that is currently being indexed:

https://github.com/zilliztech/claude-context/blob/6fc318b/packages/mcp/src/handlers.ts#L377-L392

if (this.snapshotManager.getIndexingCodebases().includes(absolutePath)) {
    if (forceReindex) {
        console.log(`[FORCE-REINDEX] Clearing stale indexing state for '${absolutePath}'`);
        this.snapshotManager.removeCodebaseCompletely(absolutePath);
        this.snapshotManager.saveCodebaseSnapshot();
    } else { … }
}

Removing the snapshot entry does not stop the background task. The old run keeps embedding and writing chunks into the collection the new run is rebuilding, and its completion handler later writes stats for a run nobody is tracking. Two writers, one collection, no ordering.

This is exactly the orphaned-writer bug that #199 fixed for clear_index — and handleClearIndex already does it right:

const activeTask = this.indexingTasks.get(absolutePath);
if (activeTask) {
    activeTask.controller.abort();
    try { await activeTask.promise; } catch { … }
    this.indexingTasks.delete(absolutePath);
}

The indexingTasks map is already populated for every background index. The force path just never used it.

Change

The force branch now cancels and awaits wind-down before clearing the snapshot entry, mirroring handleClearIndex. No new machinery, no new state.

Unchanged: force still proceeds to re-index afterwards, and a non-force call on an in-flight index still returns "already being indexed in the background" without cancelling anything.

Test

packages/mcp/src/handlers.force-during-indexing.test.ts registers a fake in-flight task whose promise only settles after its abort signal fires, then calls the handler with force: true. Asserts:

• the signal is aborted;
• wind-down completes before the handler proceeds past the already-indexing branch;
• the entry is removed from indexingTasks;
• and a second case: a non-force call reports the in-flight index and leaves the controller untouched.

It fails on master (controller.signal.aborted === false) and passes with this change. pnpm --filter @zilliz/claude-context-mcp test → 8/8; tsc --noEmit clean.