fix(tracing): report late spans under nested sent spans

[solnic]

Async work often continues from a context captured inside a nested span (a handler deep in the trace), not the root. Sent markers now cover every span of a sent transaction, and post-send storage cleanup removes sent descendants at every depth — previously only direct children were removed, leaving stale records that swallowed late spans.

Screenshots are based on an e2e scenario where finalize_batch is reported as a follow-up transaction under the nested process_batch span, starting ~0.5s after the 9ms request:

Before

Notice that finalize_batch is missing here.

After

🎉 finalize_batch is correctly under the trace.