For every variadic function, add a va_list function to the public api

[madebr]

By providing these va_list alternatives, SDL_dynapi.c can be simplified.
It should also help #9580.

What changed:
SDL_dynapi_proc.h now uses 3 macros:

• SDL_DYNAPI_PROC: behavior does not change
• SDL_DYNAPI_VPROC: variadic function returning something
• SDL_DYNAPI_VOID_VPROC: variadic function returning void

The *_VPROC macros accept 2 additional arguments: forwardfn and last_arg. This is the va_list ap function to which it should forward its arguments. The last_arg is used as 2nd argument to va_start.

I simply appended a V to the va_list alternatives.

Homework for slouken and icculus:

• come up with better names for the macros and functions
• review whether my changes to the headers are good, and orderered okay (perhaps you want to move all V declarations to the bottom)

[slouken]

My initial reaction is that it adds a bunch of entry points to the API and doesn't actually reduce or simplify much code, it just moves the complexity around. In addition, SDL_SetError() is now slightly slower, which is probably okay, but not a great selling point.

@icculus might be super happy about it though, so I'll defer to him. :)

[madebr]

These can go

[madebr]

In addition, SDL_SetError() is now slightly slower, which is probably okay, but not a great selling point.

I see the contrary on my system (which I don't understand).
Compiling the following source:

#include <SDL3/SDL.h>
#include <SDL3/SDL_main.h>
int main(int argc, char *argv[]) {
    Uint32 i;
    for (i = 0; i < 10000000; i++) {
        SDL_SetError("Time me %" SDL_PRIu32, i);
    }
    return 0;
}

and comparing times of SDL3 of current main and this pr branch, I see faster times on this pr branch.
I used "hyperfine" to do the benchmark.
I see 1.3s on current main, and 0.9s on this branch.