- use volatile for registers
- avoid implicit casts

This data corruption problem is very difficult.  Just changing the code
a little bit changes whether the problem even happens and what is
affected.  I need these changes in the main branch so I can operate with
the extra debug code always available and turned on.

Everything is setup such that if M100 is turned off or DEBUG(ECHO) is
turned off, the code is not affected.   M100 has been made a little bit
more inteligent so it can display the serial command buffers in a more
meaningful way (because the data corruption seems to often times end up
in that area).

spi.h is only needed if HEATER_0_USES_MAX6675 is defined, so making its inclusion conditional on HEATER_0_USES_MAX6675 being defined.

Configuration_store.cpp - changed a couple of print statements so the
values were printed.  The old method resulted in the formula being
printed.

ubl_G29
1) added support for R option to P4.  Now probes all points unless R is
present and the number is greater than 0.

2) P2 - moved map print statement so it showed the point that was
currently being probed,  The old location did this only for the first
point.

3) P4 - Moved the map print for the same reason.

ultralcd.cpp - changed "Mesh Editor :" to "Mesh Editor" because the LCD
draw routine puts a ":" in automatically so you end up with an extra ":"
using the old message.

- On `DELTA` the `M665 H` option supplants `M206`
- On `DELTA` `NO_WORKSPACE_OFFSETS` only reverts `G92` behavior
- Spawn 4 conditionals based on `NO_WORKSPACE_OFFSETS`
- Optimize coordinate space conversion for `DELTA` workspace
- To keep EEPROM version, retain `home_offset[XYZ]`, just ignore XY

===============================================

make changes requested by reviewers

===============================================

add M43 test to Travis, fix EOL, remove trailing spaces

Resolve 'steps_to_mm was not declared in this scope' compilation error after recent changes to merge babystepping with M851

1. The CORExx printers were checking more endstop axis than needed.

2. Removed all the CORE_xx_NOT logic.  The motor_direction(xx) routine
always returns the correct data so it is not needed.  It was actually
cause the wrong direction to be checked in some cases.

3. Made the logic/defines for X, Y & Z axis all the same.  The old logic
checked inappropriate configurations for Y and didn't check all the
correct configurations on Z.

4. Added a check for zero steps before the X, Y & Z axis.  Previously
would check the they axis even if there were no movement.

All the memory accesses need to be unsigned char in  order to avoid
false errors being detected.   Two new features are available for
developers to call into the M100 functionality to look for data
corruption.