Fix save positions

[PastorD]

Part of improving how it saves positions to avoid losing calibration

• use add_pre_shutdown_callback for more reliable saving at the end.

bump:minor

Bugfix:

• Initialize to zero fastcat::DeviceCmd cmd{}

Merge with fastcat save-positions, change to tag before merging this

@nasa-jpl/nasa-jpl-ethercat-devs

[PastorD]

• tested with latest fastca in Astrobud, no issues: move joints and move cart works well
• run a "/review" check with Opus 5 and it showed to issues. Some comments that do not neccessary

• try/catch around the call_once body — real (a throw during the destructor path means std::terminate) but not a regression; the old Shutdown() call had identical exposure.
• Calling fcat_manager_.Shutdown() instead of SaveActuatorPositions() — behavior-neutral at v0.13.20, just a coupling risk if upstream moves teardown back into Shutdown().
• The destructor deregistration + fallback themselves — unreachable today, kept as future-proofing.

[JosephBowkett]

Makes sense to me, other than catching a failed bus during motion, when we would not want to save to the normal position file, but would prefer to record the last known position in case we want to use that to recover.