Add constant-folding (simplification) of trivial cases of double-precision multiplication (MulDoublePrecXXXX)

[mostobriv]

What is the feature you'd like to have?
Constant-folding inside of IL_MUL(U)_DP, AFAIK currently it's not supported(?)

Is your feature request related to a problem?
not much

Are any alternative solutions acceptable?
Well, i can override arch plugin to change lifting in such way its won't use MUL_DP but it looks like overkill to me.

Additional Information:
Example code for arm64:

#include <stdint.h>
#include <stdio.h>


__attribute__((always_inline))
uint64_t mymul(uint32_t a, uint32_t b) {
	uint64_t result = 0;
	__asm volatile (
		"umull %x[res], %w[first], %w[second]"
		: [res] "=r" (result)
		: [first] "r" (a), [second] "r" (b)
	);

	return result;
}


int main() {
	if (mymul(5, 3) == 15) {
		puts("15 indeed");
	} else {
		puts("pls, hide me out :C");
	}

	return 0;
}

This snippet uses umull instruction, with a pretty trivial case that can be folded (two constant integers), however here is how it looks like in MLIL and HLIL:

Here is the binary i've tested on:
umull_test.zip

[galenbwill]

I have a fix for this, but it needs more testing, so I am holding it until after the Gallifrey release.

I also saw other related/similar opportunities for simplification while investigating this that I'd like to explore further.

Finally, there are some places in the MLIL data flow query system that can benefit from similar changes, that will improve value set prediction, which I will include with this fix.