This works around a bug in Pico SDK. A poorly-timed short sleep can
disrupt all timers in the default alarm pool. There seems to be a race
condition - the ALARM3 register is probably set to a value in the past.
This causes the alarm to never fire anymore even though the alarm appears
to be armed.

This workaround forces the ISR to run in this might have happened.

Related to https://github.com/raspberrypi/pico-sdk/issues/737

Using malloc in interrupts is a bad idea. This change
instead uses a preallocated buffer for all modes.

In its current state, most of the typedata is also ignored
to not waste memory.

This was a *bad* idea and I was lucky that it did not explode.

Adjust RP2040 ADC Vref constant to make battery voltage more accurate

See merge request !3

My brick was previously under-reporting the measured voltage.
To investigate this, I have probed the PCB and got these results:

battery voltage    ... 8.25 V
voltage at ADC pin ... 2.06 V (OK, voltage is correctly divided by 4)
ADC ref. voltage   ... 3.2 V
ADC measurement    ... 42122 (OK, corresponds to cca. 2.06 V)

Battery objects can be fully self-contained -- the ADC API
that the RP2040 SDK provides can be used from multiple objects
at once.

This can be contrasted with the encoder implementation --
the interrupt processing there must have some global state
(be it the position counters themselves or a list of
interrupt-handling objects).

This should make garbage collection of EV3 sensors/motors slightly
more reliable.

This prevents a bug where encoders do not work after a soft reboot.
This reboot disabled pin IRQs, but it did not tell the encoder
module to reenable them on reboot.

This change instead relies on Motor's __del__ being called before
soft reboot. This way the encoder module can be gracefully deinitilized
before the reboot happens.

However, as the encoder module is shared between three consumers
(jv_motors, opencube_motors, Encoder object), additional refcounting
is needed to avoid unwanted interactions between them.

This is a sneaky way of disabling speed ramping - infinite acceleration
will result in zero rampup/rampdown times.

The previous default led to a slight, but audible motor shaking
when a command like run(0) was used.

This should also decrease the action signal "noise" - previously
the signal looked like a sawtooth as the measured motor position
changed only discretely.

This will limit the minimum measurable speed to ~50 deg/s, but it should
reduce the severity of "ghost measured speed" after quickly stopping
the motor.

These constants were generated with Matlab pidtune with 'DesignFocus'
set to 'disturbance-rejection'. With feedforward in place, feedback
can focus on eliminating effects of external disturbances.