Macroscope caught a real regression in the first pass: raising the
shared SSH_READY_PROBE_TIMEOUT_MS to 8s also fed the remote-side
reuse/launch scripts' own loopback probe. That script's deadline loop
only checks its overall timeout *between* attempts, so one in-flight
probe bounded by 8s could blow through the reuse check's 2s budget by
several seconds — a real regression the previous version didn't
account for.

Split it: SSH_READY_PROBE_TIMEOUT_MS stays 1s for the remote script
(always loopback, no RTT to account for, and must stay under
REMOTE_REUSE_READY_TIMEOUT_MS), and a new
SSH_TUNNEL_READY_PROBE_TIMEOUT_MS (8s) applies only to the desktop-side
probe that actually crosses the SSH link at initial connect. That path
uses Effect's own timeoutOption, which genuinely interrupts an
in-flight request at its outer deadline, so it can't overshoot a short
timeoutMs elsewhere the way the remote script's loop can.