fix(node-sdk): await in-flight bulk requests when flushing on Lambda

[roncohen]

Summary

• Track and await already in-flight bulk deliveries during flush(), including timer-triggered and full batches.
• Add regression coverage for overlapping batches and failed delivery.
• Document Lambda use with ReflagClient, request-driven refresh, and awaited flushing in finally.
• Add a SAM-deployable Lambda lifecycle probe using local SDK source, with delayed local HTTP and optional real-endpoint modes.
• Include a Node SDK patch changeset.

Why

Previously, an automatic flush cleared the buffer before delivery completed. A subsequent explicit flush could return immediately, allowing Lambda to freeze with bulk I/O still pending. Workers use waitUntil; Lambda handlers must await flush before returning.

This fixes a demonstrated lifecycle race, but is not yet confirmation of the reported customer timeout root cause. Both ReflagClient and EdgeClient share the affected buffer.

Validation

• Before rebasing onto current main: all 236 Node SDK tests passed; SDK build and strict probe type-check passed.
• Local HTTP A/B probe: original SDK returned with one request pending in both timer/full modes; patched SDK returned with zero pending and one successful delivery.
• git diff --check passes.
• Post-rebase test rerun blocked by dependency installation: main upgraded Vitest; yarn install did not complete the fetch phase in this environment. CI should validate the current base.
• Actual AWS deployment, warm/cold freeze-thaw testing, and real-endpoint latency measurements are still pending sandbox credentials/region.

Scope

No automatic Lambda detection, timeout increases, external queue, or extension. The documented pattern awaits flush on every invocation; delivery failures remain logged and discarded.