fix(platform-node): preserve queue and IPC ordering

[LukasParke]

What

• keep durable queue sequence numbers monotonic across clear()
• delete acked frames by computed sequence key instead of scanning storage
• serialize inbound IPC frame dispatch per client
• correlate server error frames to the send/ack they reject
• cap socket writable backlog for stalled consumers
• use Linux /proc pid start-time when available and support async process identity checks

Why

Resetting the durable sequence head could cause a resumed peer to discard new frames as already acknowledged. Frames sharing a TCP chunk could execute out of order, and one uncorrelated error rejected every pending send. Process identity checks also spawned ps repeatedly.

This semantically ports the queue/IPC/subprocess portion of fork/port/openrouter-fixes commit b9905ac5. File-storage behavior is separately reviewed in #70.

Test plan

• platform-node suite — 64 pass, 1 benchmark skip
• core/platform-node typechecks
• ordered burst and correlated-error concurrency tests
• root lint
• sentrux check .
• sentrux gate .
• clean-context Pi review

Known boundary

This PR preserves existing live/replay socket semantics; fully serializing concurrent replay and live producers is a separate transport design change rather than part of the historical port.

[vercel]

@LukasParke is attempting to deploy a commit to the Matt Apperson's projects Team on Vercel.

A member of the Team first needs to authorize it.

[LukasParke]

Agent: Code CI/structural/DCO are green where configured; remaining external red checks are compat with an empty upstream OPENROUTER_API_KEY and Vercel authorization. Where core CI is red, the only failure is the unrelated timing-sensitive executeRunCode retry-backoff test; affected local/full suites passed. I cannot rerun upstream Actions without admin rights.