[c++] Add static initialization guards for Cortex-M

[salkinium]

This adds a hopefully atomic statics initialization guard implementation to implement #345.
In case of recursive initialization, a debug assertion cxa.guard.recursion is raised.
For AVRs, the threadsafe statics are not implemented at all to save code space.

Should the guard implementation be optional for ARM Cortex-M? The guard itself takes 4 bytes per function static, and the guards a little bit of code. I don't think this is too much size, but for very small targets it may be useful to have this as an option. I'm also not aware of this not being atomic being an issue in the past.

Not sure how to overwrite this for FreeRTOS use, I'll fix it when it becomes an issue.

cc @chris-durand @rleh @dergraaf

[salkinium]

I also had to implement the GCC atomic builtins for ARMv6-M, since apparently the toolchain delegates the implementation to the user for these things. They are now done using atomic locks, which is fairly brute force, but should work just fine.
https://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html