Also...   The memory corruption issue may be fixed.   The GCC compiler
was inlining static functions and this caused the G29() stack frame to
become much larger than the AVR could handle.

* relocated ubl state to config. store:
 * removed a number of ubl state variables and padding which were largely unused - saved 58 bytes of both SRAM and EEPROM;
 * modified ubl sanity_check - no longer checks removed state variables that were otherwise unused, where checking didn't seem to accomplish anything, ultimately;
 * removed pre_initialized state, saving 64 bytes of SRAM;
 * removed automatic saving of UBL state after UBL activation/deactivation;
* consolidated multiple GRID_MAX_POINTS_X/Y to 'Global Leveling' section of EEPROM;
* minor update to G29 Sx notes/instructions;
* renamed mesh load and save parameter to 'slot' from 'm' for clarity;

fix maximum mesh_index_ array size at 16 (15+1);

Pretty much...  The code is in place.  Still more work to do.    But it
has a lot of hooks and variables in other code, so commit and merge
before I pick up a million 'Conflicts'.

As it turns out...   an unsigned int can not go to less than 0,  so the
loop never terminates.

- Make all `unified_bed_leveling` data/methods static
- Move some UBL-related variables into the class
- Replace `map_[xy]_index_to_bed_location` with `mesh_index_to_[xy]pos`

UBL's maps are cleaner.   And with a 1 value, G29 O 1  will generate a
map suitable to be Cut & Pasted into Excel so a Suface Map can be
generated.

The wait_for_user change totally broke UBL.   But there is stuff wrong
now in the thermal code and/or LCD Panel code.