fix(plugin): async shutdown hook returns forceFlush Promise + drop sdk.shutdown bus race

[Alezander9]

rc3 diagnostic captured two compounding bugs.**1.** bcode-laminar's server.instance.disposedbus handler awaitssdk.shutdown(), which unregisters the global TracerProvider AND closes the BatchSpanProcessor. In headless bcode runmode this fires before the top-levelfinally, so by the time the sync shutdown hook runs, the provider is gone and the BSP queue is dropped. Diagnostic: [bcode-laminar] shutdown invoked: ending 0 turn span(s).2. The top-level finally fell back to trace.getTracerProvider().forceFlush(), but the OTel API's ProxyTracerProvider does not expose forceFlush — so even when the global was still registered the fallback was a no-op. Diagnostic: [bcode] shutdown: no forceFlush on global provider.**Fix.**``- Hooks.shutdownreturn type widened tovoid | Promise so plugins can return a Promise the host awaits.- bcode-laminar.shutdown now ends any still-open turn spans then returns processor.forceFlush() directly. The closure ref to processor bypasses the broken global lookup entirely.- Drop await sdk.shutdown()fromserver.instance.disposed (the handler now just ends remaining spans).- src/index.ts: drop the trace.getTracerProvider().forceFlush() block and the now-unused trace import. Host awaits Promise.allSettled of hook returns with a 3s race.Diagnostic process.stderr.writelines stay so the next cloud verification can confirmforceFlush done in Xms. They will be removed in a follow-up before v0.1.8 proper.Typecheck: 7/7 packages green.

---

Summary by cubic

Fixes dropped OTel spans during shutdown by making plugin shutdown hooks async and removing the race caused by early SDK teardown. The host now awaits plugin flushes with a 3s cap, and bcode-laminar flushes its own processor reliably.

• Bug Fixes
  • shutdown hook now supports void | Promise<void>; host awaits all hooks with a 3s timeout.
  • bcode-laminar shutdown ends open turn spans, then returns processor.forceFlush() via closure (no global provider lookup).
  • Removed sdk.shutdown() from server.instance.disposed to avoid unregistering the provider and closing the BSP prematurely.
  • Dropped trace.getTracerProvider().forceFlush() fallback and the unused trace import; the proxy provider doesn’t expose forceFlush.
  • Kept temporary stderr diagnostics to verify flush timing; to be removed later.

^{Written for commit a6e7a87. Summary will update on new commits. Review in cubic}