Unit test for LoadWordLeft

[querolita]

This PR provides a unit test for the `LoadWordLeft` instruction. It matches closely the Cannon definition of the instruction given by

		case 0x22: // lwl
			val := mem << ((rs & 3) * 8)
			mask := uint32(0xFFFFFFFF) << ((rs & 3) * 8)
			return (rt & ^mask) | val

I used this test to realize that our current implementation of the interpreter for that function was not working properly. In particular, bytes were being mapped to different positions. I changed the interpreter code to avoid the confusing (and error prone) use of overwrite_n in favor of a more direct set of variables called mod_n. I included comments to help understand what the above function does and what it expects in each byte position of the destination register.

Old description

This unit test matches what the interpreter of LoadWordLeft is doing. Nonetheless, the constraints still do not pass when trying to make a proof for them. In particular:

"Unsatisfied expression: ((((Curr(x[19]) - 0x0100000000000000000000000000000000000000000000000000000000000000) * Curr(x[23])) + Curr(x[22])) - 0x0100000000000000000000000000000000000000000000000000000000000000)"

Still figuring out why. The obvious answer would be dummy rows initialized to zeros so the -1 will not hold. But we already solved that issue when creating the trace, by padding with a dummy row that is known to be a "real" row.~

[querolita]

After applying #2311 to this PR, I will run the nodes again to see if the constraint holds now.

[querolita]

Demo halts, making the PR a draft until solved

[querolita]

Closing due to #2729