MicroPython: Magic C++ Memory Management.

[Gadgetoid]

This is more a test to see if I'm barking up the right tree than a fix... yet.

[Gadgetoid]

Okay as a result of sleuthing for #720 I ended up raising micropython/micropython#11116

After raising my concerns in micropython/micropython@c80e7c1

And spending a large amount of time trying to avoid new and delete.

Unfortunately, C++ does a lot of heap allocation that's difficult to control. In fact our project does ~10k and it's a job for Future Phil to figure out why on earth we're using so much heap because that seems extreme.

As a result of hours upon hours of tinkering I finally came up with https://github.com/Gadgetoid/micropython-cppmem which is integrated directly into pimoroni-pico in this PR.

This new module just needs to be linked into a MicroPython build and it takes over all C++ memory allocations in two ways:

1. All pre-init allocations end up in a 10k, linear, non-freeing memory pool that survives until hard-reset
2. Once init is done, the module switches into "MICROPYTHON" mode, where new and delete map to m_malloc and m_free respectively.

A side-effect of this approach is that any memory resulting from an implicit or explicit use of new in C++ modules cannot be relied upon to not be garbage collected.

This is to say that all our best practices still apply- if you're allocating something that you want to last for more than one C++ function call then it needs to be allocated with m_new and a pointer to the resulting memory region must be stored on an active MicroPython object. Like how our classes work with pointers to buffers and i2c objects.

[Gadgetoid]

The builds breaking and failing to start up when the cppmem/micropython module is included before other modules suggest that __attribute__ ((init_priority (65535))) is not global. If this is the case, then moving the wakeup module further up the list of C++ modules could make Badger 2040 / Badger 2040 W wakeup button detection a little more responsive.

[Gadgetoid]

Should probably move the switch to MicroPython heap mode into boot.py since we can be sure it will always run after the MicroPython heap, and C++ objects, have initialised.