nrf: neopixel_write: busy poll for one pixel writes

It's probably not the whole story, however, this fixes a crash observed
when bulk copying data to an nRF board using `dd`.

Basically, the call stack looked like this when resetting into safe mode:
    #0 reset_into_safe_mode reason=reason@entry=GC_ALLOC_OUTSIDE_VM
    #1 gc_alloc
..  #4 external_flash_write_block
.. adafruit#11 usb_background
   adafruit#12 run_background_tasks
   adafruit#13 common_hal_neopixel_write
.. adafruit#18 start_mp

i.e., during early startup, it is not okay yet to call allocation functions
like m_malloc_maybe that use the garbage collected heap.  However,
nRF's neopixel_write (which already includes special handling to avoid
heap allocations for the status pixel!) can enter background tasks, which
do nearly arbitrary things including heap allocations.

We re-use the same test that switches from heap allocation to stack
allocation for the pattern buffer.

Author: jepler

ports/nrf/common-hal/neopixel_write/__init__.c

@@ -200,7 +200,9 @@ void common_hal_neopixel_write (const digitalio_digitalinout_obj_t* digitalinout
 
         // But we have to wait for the flag to be set.
         while ( !nrf_pwm_event_check(pwm, NRF_PWM_EVENT_SEQEND0) ) {
-            RUN_BACKGROUND_TASKS;
+            if (pattern_size > sizeof(one_pixel)) {
+                RUN_BACKGROUND_TASKS;
+            }
         }
 
         // Before leave we clear the flag for the event.