Return free/allocated memory on GC heap.

msvc: Enable GC

It sucks to workaround this on uPy side, but upgrading not upgradable
embedded systems sucks even more.

So, allocate one explicitly.

The pointers to the bss section are acquired in init.c()
by inspecting the PE header. Works for msvc and mingw.

These changes were tested with QEMU, and by few people of real hardware.

It defines types used by all other headers.

Fixes #691.

In generalize case, optimization options should be passed to all stages of
the build process.

Expected to be set on command line, with the idea being that for different
targets, there're different smartass ABIs which strive to put unneeded
sections into executables, etc., so let people have flexible way to
strip that.

The option name is similar to previously introduced CLFAGS_EXTRA &
LDFLAGS_EXTRA.

The idea is that it should be possible to pass any additional params for
experimentation without need to patch sources (and without need to deviate
from or repeat baseline options).

Some people want to enable even more warnings. Let them do it without putting
burden on everyone. Some people vice versa think that current settings should
be relaxed. In this regard, -Werror is the most problematic, it disallows to
use #warning directive, and disallows to pass configuration settings on make
command lines. Again, until decided how to deal with these globally, allow to
work around these problems locally.

One thing is wanting to do 1 / 2 and get something else but 0, and quite
another - doing rocket science ;-).

gc: More verbose debugging

Fix missing declaration of assert()
Replace ARRAY_SIZE with MP_ARRAY_SIZE

map() is 5 times slower. That's mostly because of inefficiency of creating
containers from iterables of unknown length (like map()).

Both "bound" (like, length known) and "unbound" (length unknown) are tested.
All of list, tuple, bytes, bytesarray offer approximately the same
performance, with "unbound" case being 30 times slower.

For a trivial operation, calling a function is 5 times slower than doing
operation inline.

bare-arm, stmhal: Disable stack protector

As we are building with -nostdlib gcc features like the stack protector
will fail linking, because the failure handlers are in gcc's internal
libs. Such features are implicitly disabled during compilation when
-nostdlib is used in CFLAGS too.

Signed-off-by: Sven Wegener <sven.wegener@stealer.net>

-nostdlib is the correct option, gcc recognizes the double dash version
when in link-only mode, but not when compiling.

Signed-off-by: Sven Wegener <sven.wegener@stealer.net>

Fix problem with ADC reads and multiple channels

Add more DEBUG_printf statements to trace gc behaviour