Momentary Switch Homing and Dome Rotation Calibration

[CTetford]

Adds homing support for momentary (magnetic/limit) home switches, which are only ON for a short time. The stock fast/slow/fine refinement has no region to converge on with these. All opt-in; default behaviour is unchanged.

Momentary homing

• (HOME_SEQUENCE_MOMENTARY): seek past the switch until it releases, approach from the far side, record the coordinate at the sensor trip, then goto back to it.
• Moves are separated by a settle delay (HOME_SETTLE_TIME), edges are debounced by travel distance (HOME_SENSE_DEBOUNCE), and seek direction is selectable (HOME_SEEK_REVERSE).
• The dome now sets home from the recorded trip point, and only when the sensor was actually reached - a timeout or motor fault won't write its stopping point to NV as home.

:DM# rotation calibration

• Rotates continuously, records each home switch pass, and reports averaged steps/revolution for calibrating AXIS1_STEPS_PER_DEGREE. Requires AXIS1_WRAP ON; writes nothing to NV; :DH# stops it.

Servo safety - oscillation detection update

• Old: shut down if drive output went below -33% and above +33% within the same 2-second window. This approach caused nuisance trips on normal behaviour: a commanded direction reversal, or the settling transient after a stop.
• New: Hägglund's method (https://doi.org/10.1016/0967-0661(95)00164-P) applied to the following error: integrate |error| between zero crossings, count a half cycle as significant only if its mean magnitude and duration are both in range, and act only on consecutive significant half cycles. A one-off transient scores once and resets. Warns at half the shutdown count and reports the measured period and amplitude. Tunable via SERVO_SAFETY_OSCILLATION_*.

Servo safety - runaway detection update

• Old: shut down if the distance travelled in a 2-second window exceeded the previous window's target distance while at ≥90% power. Comparing distances means a following error that just grows (heavy load, speed change) can look like a runaway, and a real runaway can hide.
• New: shut down only if the encoder moved in the opposite direction to the target at ≥90% power, on two consecutive checks. Direction is unambiguous where distance is not, and the two-check requirement keeps a window containing a commanded reversal from qualifying. Still catches a reversed encoder.

Fix - backlash takeup deadlock

• When a slew reversed direction with backlash compensation set, the axis stopped dead partway through the move and never recovered. No error, no timeout.
• Cause: while the motor is taking up backlash, Axis::poll() commanded a frequency of zero. The motor backends read travel direction from the sign of the commanded frequency, and zero has no sign - so the motor stepped nowhere, the takeup counter never advanced, and inBacklash never cleared. The fix keeps the normal frequency ramp running during takeup; the backends already clamp it to the backlash rate, so takeup completes and the slew continues. Dome homing also zeroes backlash for the run (as park/unpark already do) and restores it on every exit path.

New options (HOME_SEQUENCE_MOMENTARY, HOME_SEEK_REVERSE, HOME_SENSE_DEBOUNCE, HOME_SETTLE_TIME, DOME_CALIBRATE_REVOLUTIONS, SERVO_SAFETY_OSCILLATION_*) all default to current behaviour and are checked in Validate.h.

Tested on a modified Technical Innovations HD6 dome, Teensy 4.1 / OCS4: SERVO_EE DC drive, AB encoder, magnetic home switch.