compiler,runtime: implement stack-based scheduler

[aykevl]

This scheduler is intended to live along the (stackless) coroutine based
scheduler which is needed for WebAssembly and unsupported platforms. The
stack based scheduler is somewhat simpler in implementation as it does
not require full program transform passes and supports things like
function pointers and interface methods out of the box with no changes.

Code size is reduced in most cases, even in the case where no scheduler
scheduler is used at all. I'm not exactly sure why but these changes
likely allowed some further optimizations somewhere. Even RAM is
slightly reduced, perhaps some global was elminated in the process as
well.

[aykevl]

Note that this will conflict with #487. Either could be merged first, I'll fix the conflicts in the other PR.

[aykevl]

This PR sets a fixed goroutine stack size of 1kB by default. I am not sure what a good default size would be, that should probably be made configurable at some point. But a default of 1kB may be good enough for most embedded systems. Also: TinyGo likely requires less stack size anyway than the main Go implementation (untested!) because of the different calling convention and the fact it is usually used for 32-bit systems (halving int and pointer sizes).
There is a check in place that should catch most stack overflows when switching between stacks, but other than that a stack overflow may corrupt memory at the moment. I have a long-term goal of using the .stack_size section to determine the stack size whenever possible at compile time and inserting runtime checks when that is impossible, but that will require a decent amount of work.

This new goroutine implementation can be switched off by passing the -scheduler=coroutines option. The default for Cortex-M is -scheduler=tasks and elsewhere -scheduler=coroutines.

[deadprogram]

I can confirm that this PR does address the issue I encountered with my use of multiple go routines.

Merging this PR, so that #487 can have the conflicts resolved.

[deadprogram]

OK spoke too soon. It did solve my compilation program, but then code did not operate.

I then went and looked more closely, and discovered that the USB CDC code on the SAMD21-based boards does not function after this commit.

This code no longer works, specifically the USB port does not appear on my machine as expected:

tinygo flash --target arduino-nano33 examples/serial

However this works:

tinygo flash -scheduler=coroutines -target arduino-nano33 examples/serial

Would appear to be the new scheduler...

[aykevl]

Ouch, I should have investigated why there was a code size change at all. Because I probably introduced a bug somewhere that led to LLVM optimizing something it shouldn't.

I will investigate tomorrow.

UPDATE: see #507