After wraparound, pwm_count <= pwm_mask holds, thus soft_pwm_X <= pwm_count
guarantees soft_pwm_X < pwm_mask is true, and the heater will be switched
off in the first branch.
Do not evaluate the pwm conditions a second time, this reduces the
instruction count (4 instructions per PWM) and text size (6 byte).

Signed-off-by: Stefan Brüns <stefan.bruens@rwth-aachen.de>

The compiler is not able to reuse the value of pwm_count, but reloads it
on every evaluation, if is stored in a static variable, as it cannot prove
it will be unchanged. A variable with local scope may not be modified from
the outside, so its value can be reused.
Doing so reduces text size and instruction count.

Signed-off-by: Stefan Brüns <stefan.bruens@rwth-aachen.de>

If dithering is enabled, the remainder of the soft_pwm_X duty value at
turnoff time is added to the next cycle. If e.g. the duty is set to 9 and
SCALE is set to 2, the PWM will be active for 8 counts for 3 cycles and
12 counts on each fourth cycle, i.e. the average is 9 cycles.

This compensates the resolution loss at higher scales and allows running
fans with SOFT_PWM with significantly reduced noise.

Signed-off-by: Stefan Brüns <stefan.bruens@rwth-aachen.de>

A 128 step PWM has 127 intervals (0/127 ... 127/127 duty). Currently, a
PWM setting of 1/127 is active for 2/128, i.e. double the expected time,
or, in general n+1/128 instead of n/127.
Fixes issue#6003.

Signed-off-by: Stefan Brüns <stefan.bruens@rwth-aachen.de>

The code to print a hex nibble got broke.   SERIAL_ECHO( (char) ) isn't
suitable for its use.

All the changes to get the code into compliance with Marlin coding
standards made UBL sick.   This pretty much gets UBL back up and
running.

Especialy against stepper ISR.
This is even more important when a minimum pulse width is set,
increasing the runtime of a babystep.

Sometimes after killing heaters are still on.

G29 P4 R O  should work now.

The retraction flag in UBL's G26 was conflicting with the FWRETRACT option.

All fixed!

Fix the logic change that happened when UBL got merged.

Thanks to @ospalh -- #6056