Fix: increased debounce delay.

With the introduction of the #if ENABLED(…)

reads better than

* Updates the auto bed leveling comments ONLY.
* Renames "Z probe" word usage in //comments and #errors properly.

Always end at [RIGHT_PROBE_BED_POSITION, BACK_PROBE_BED_POSITION]
Regardles the evennes of auto_bed_leveling_grid_points.

Use the shortest way between probing-points.

and use it.

Currently we use the probe exclusively as a device to find the build platform(bed).
For the currently supported setups this means, we use it as a additional min-endstop.
A triggered when not deployed probe disturbs the homing process for max-endstops.

Rename ENDSTOPPULLUP_ZPROBE to ENDSTOPPULLUP_ZMIN_PROBE
Rename Z_PROBE_ENDSTOP_INVERTING to Z_MIN_PROBE_ENDSTOP_INVERTING
Rename Z_PROBE_ENDSTOP to Z_MIN_PROBE_ENDSTOP
Rename DISABLE_Z_PROBE_ENDSTOP to DISABLE_Z_MIN_PROBE_ENDSTOP
Rename Z_PROBE_REPEATABILITY_TEST to Z_MIN_PROBE_REPEATABILITY_TEST
Rename Z_PROBE_ENDSTOP to Z_MIN_PROBE_ENDSTOP
Adjust comments accordingly
Remove Z_MAX  check for the probe in update_endstops().

Using an delta related idea of @clefranc from #61,
extended to the general change for all setups.

Tested with Prusa i3, max-z-endstop and permanently triggered z-probe.
Worked for @clefranc's delta.

If the printer is going to accept the M999,
then it should continue from that line number.

The long descriptor will still be available in M115

Saves about 473 byte of progmem and 28 bytes of RAM.

Just the minimal changes to qr_solve.cpp and .h to get rid of malloc() and free().

Saves about 656 bytes of progmem (library-code)
and 22 bytes for static variables.

Should use exactly the same amount of stack as it did before on the heap.

Include all the necessary pieces to get the final Z position after
probing the bed, and add commentary about other possible changes.

Updates to `set_homing_bump_feedrate`:
- Move the string into Program Memory, reduce length by 31 bytes
- Use an auto to get the divisor, adjust it on error
- Set feedrate once, at the end

The compiler may be able to optimize if it knows an array won’t be
changing.

`gcode_M84` only needs to call `finishAndDisableSteppers`, which calls
all the other functions.

- Update prepare_move_delta to take a target argument
- Add Delta support to plan_arc

- Have `Servo::attach` explicitly return -1 if it fails
- Check for -1 in `Servo::move` because `servoIndex` might be 0
- Make `attach` / `detach` calls conditional on `SERVO_LEVELING`
- Move `SERVO_LEVELING` define to `Conditionals.h`