enable TCP keepalive, and replay in-flight resends on every reconnect

[tirewapped]

fix: replay in-flight resends on every reconnect

Replay was gated on the daemon instance token changing. On a same-instance
drop, a request written into a socket that died before the daemon read it
was never delivered again; the caller's _send_and_wait blocked forever,
wedging the client. The tracked requests are keyed by envelope hash and the
daemon handles re-issues idempotently (BoxAlreadyExists on writes, re-read
of the same box on reads), so replaying on every reconnect is safe. Adds
worker-loop unit tests pinning this behavior.

fix: enable TCP keepalive + user timeout on daemon socket

A silently dead peer (no FIN/RST) left recv() blocked forever, so the
worker loop never detected the drop and never replayed in-flight requests.
Set SO_KEEPALIVE with 10s/10s/3 probes and TCP_USER_TIMEOUT 30s so the
kernel aborts a dead link and surfaces an error to the read loop.