v0.57.0

Fixed

• The Codex tmux wake transport now types the wake prompt and presses Enter as
  two separate tmux send-keys calls with a configurable --submit-delay
  pause. A single combined call left the prompt unsent in the Codex input
  buffer, so injected wakes were silently dropped until a human pressed Enter.
• synapse codex-tmux wait no longer exits on the first failed synapse wait.
  It retries with capped exponential backoff and only gives up after
  --max-wait-failures consecutive failures (unbounded by default), so a brief
  hub restart or eviction no longer kills the waker permanently.
• Connection failures from the command-line verbs now distinguish a hub that is
  full, in a takeover cooldown, or rejecting a duplicate name from a hub that is
  simply absent. A full hub previously printed the same Could not reach hub
  line as an offline one, masking a capacity ceiling as an outage.
• synapse git-claim, synapse lock, and synapse release now name the real
  reason a request got no reply instead of printing no response from hub or
  timed out. Claiming or locking under a name another live session already
  holds — a common slip when reusing a waiter's identity — now reports the name
  conflict (close 4009), and a full hub reports its capacity (close 4013). They
  also stop waiting as soon as the hub closes the socket rather than polling out
  the full window.

Changed

• Raised the default hub connection ceiling (--max-clients) from 64 to 256 so
  a multi-project fleet, where each terminal holds a command socket and a
  persistent waiter, does not exhaust the table and reject new connections with
  close code 4013.