fix: bound beacon runtime and sync work

[lodekeeper-z]

Summary

• subscribe to data-column subnets derived from the node's actual custody columns
• bound req/resp protocol opens and the complete response transfer so slow peers cannot monopolize the P2P loop
• apply BLS dispatch gates to recovered unknown-block gossip work
• preserve peer ownership across in-flight sync downloads and keep validation/rewind paths within the pending batch window
• exclude the temporary QUIC/gossipsub tracing dependency, debug metrics, trace scripts, and investigation-only artifacts

Review notes

• Response timeouts use one absolute deadline for the complete multi-chunk transfer, matching Lodestar-TS respTimeoutMs semantics rather than resetting per chunk.
• Sync retry dispatch remains capped after a long awaiting-validation prefix is rewound.
• Batch.startDownload is transactional under allocator failure: no state, generation, metric, or callback changes occur unless peer ownership succeeds.

[spiral-ladder]

you're alive!

[lodekeeper-z]

Alive and catching up 😄