HistoryJournalStore.dispose() is fire-and-forget: recorder dispose errors are swallowed and scratch is removed asynchronously #3867

Problem

In PR #3852 (#854), HistoryJournalStore.dispose() (packages/core/src/services/history/historyJournalStore.ts) is synchronous. When the store owns its recorder, it hands the real work to a detached promise chain:

void this.fanoutFeed
  .settle()
  .catch(() => undefined)
  .then(() => recorder.dispose())
  .catch(() => undefined)
  .then(() => {
    if (this.binding.ownsRecorder) removeTempDir(tempDir);
  });

That has three consequences:

1. Errors disappear. A failure to settle mirror delivery or to dispose the recorder is swallowed twice, so nothing reports it. This conflicts with the project's fail-fast rule.
2. Callers can't wait. Nothing can await completion, so code that disposes and then exits, or disposes and immediately checks scratch, races the cleanup.
3. Scratch can leak. If the process exits before the chain runs, the llxprt-history-* scratch directory stays on disk.

Proposal

Make disposal awaitable, for example dispose(): Promise<void> or an [Symbol.asyncDispose], and let failures propagate to the owner that disposes the store. Keep a synchronous path only where an owner genuinely cannot await, and make that path report failures somewhere visible instead of discarding them.

Acceptance

• A recorder dispose failure surfaces to the disposing owner.
• After awaited disposal, the store's scratch directory no longer exists.
• Existing scratch-after-dispose and journal lifecycle tests stay green.

Follow-up from #854 / PR #3852, listed there as a known open item.