Allow 5 seconds for an Archon reply

[mikelangmayr]

• Every command waited 3000 ms for a reply, but APPLYALL measures 4.5 to 4.8 s on an Archon whose bias complement makes its power-down ramp slow, so load could never succeed there
• A timed-out command is worse than a failed one: its reply arrives afterwards and is read as the next command's, leaving the link one reply behind until it is reopened
• Verified against that controller through the Python bindings: load now completes where it previously failed at 3.002 s, and bias, heater and repeated STATUS reads work on the same connection afterwards