Always honor minimum period on stepper pulse generation, and fix timing calculations

Signed-off-by: etagle <ejtagle@hotmail.com>

* Misc fixes and improvements

- Get rid of most critical sections on the Serial port drivers for AVR and DUE. Proper usage of FIFOs should allow interrupts to stay enabled without harm to queuing and dequeuing.
  Also, with 8-bit indices (for AVR) and up to 32-bit indices (for ARM), there is no need to protect reads and writes to those indices.
- Simplify the XON/XOFF logic quite a bit. Much cleaner now (both for AVR and ARM)
- Prevent a race condition (edge case) that could happen when estimating the proper value for the stepper timer (by reading it) and writing the calculated value for the time to the next ISR by disabling interrupts in those critical and small sections of the code - The problem could lead to lost steps.
- Fix dual endstops not properly homing bug (maybe).

* Set position immediately when possible

- Also implemented real endstop reading on interrupt.

Disabling an ISR on ARM has 3 instructions of latency. A Memory barrier is REQUIRED to ensure proper and predictable disabling. Memory barriers are expensive, so avoid disabling if already disabled (See https://mcuoneclipse.com/2015/10/16/nvic-disabling-interrupts-on-arm-cortex-m-and-the-need-for-a-memory-barrier-instruction/)

Use macros that explicitly avoid double-evaluation and can be used for any datatype, replacing `min`, `max`, `abs`, `fabs`, `labs`, and `FABS`.

Co-Authored-By: ejtagle <ejtagle@hotmail.com>

Also simplify logic on all ARM-based interrupts. Now, it is REQUIRED to properly configure interrupt priority. USART should have highest priority, followed by Stepper, and then all others.

Followup to #9985

To reduce confusion over the current timer count vs. the compare (aka "top") value. Caution: this re-uses the function name, changing its meaning.