Mako: recover from a lost camera link without restarting noctua-app

A VmbError.InternalFault (link drop on the telescope LAN) left the dead
camera handle cached forever: every call failed on it, errors piled up
in device.error (2 entries per telemetry poll) and only a process
restart brought teccam2 back.

- mako.py: drop the handle on link-loss VmbC codes (InternalFault,
  BadHandle, DeviceNotOpen, ApiNotStarted) and reconnect on the next
  call; error list holds only the current problem; retry interval after
  a failed connect; recover() reopens the camera, then restarts the
  VmbSystem as a last resort; the loop reconnects on its own and
  reapplies exposure on every new connection; connection property,
  download() without frame, DATE-OBS on retries, binning read once per
  loop connection instead of once per frame.
- check.vmbpy_errors: record once via device.fail(), clear on success.
- api Loop.post: run off the event loop, re-arm the camera first.
- loop-toggle.js: switch follows response.running, not the HTTP status.
- guider.py: same structure for the two telescope offset blocks.
- Rename the teccam device classes stx.Guider/mako.Guider to Probe,
  with their routes in the new noctua.api.probe.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>