winapi: Run fls_init in startup thread, before regular constructors

[thrimbor]

This, together with XboxDev/nxdk-pdclib#48, fixes the issue where constructors were run before thread-local storage was available by running them from the main thread instead of the startup thread. However, fls_init is a special constructor that needs to run before any thread creation, so this introduces and uses a new mechanism which allows for constructors to run in the startup thread instead by putting function pointers in a custom subsection of .CRT (the same thing is done automatically by the compiler for regular constructors).

This code can be used to verify that the change works. Before the changes, this would crash during the construction of the global std::stringstream s, but with the changes, it should run correctly.

#include <hal/debug.h>
#include <hal/video.h>
#include <windows.h>
#include <sstream>

std::stringstream s;

int main(void) {
  XVideoSetMode(640, 480, 32, REFRESH_DEFAULT);

  debugPrint("hello\n");

  while (true) {
    Sleep(2000);
  }

  return 0;
}

Fixes #496.

This will be combined with a submodule update and turned into a normal PR when XboxDev/nxdk-pdclib#48 is merged.

[thrimbor]

This has sat for a while, and as it is a crash fix, I'm self-merging this now.