progress-aware mount startup

[Prasanna721]

mount used to bail at a fixed 30s if the daemon didn't come ready,
which broke big containers where initial sync legitimately takes
longer. now the parent watches a per-tag startup.json the daemon
updates through each phase (validating key, opening cache, initial
sync, mounting fs, starting ipc, ready) and only times out after
30s of no progress. flag is --startup-timeout, default still 30s.

initial sync still runs before the mount goes ready, but the parent
now shows a smoothed X / Y files loaded line on a tty while it
waits. sync engine grew _with_progress variants for deletion scan
and full pull so the daemon can pump counts up.

also: parent uses child.try_wait() instead of pid-alive polling, so
a daemon that exits early surfaces the actual exit status, and
stale startup.json files get cleaned up on shutdown and on
stale-tag sweeps.

how I tested:

• 3 new unit tests around StartupWaitState (timeout fires after
  inactivity, progress resets the timer, identical bytes don't
  reset it)

[vorflux]

Two actual bugs found — see inline comments.

[linear]

ENG-699 Review PR #12 in smfs