Add option to treat literal assignments as operations in SSACFGs

[clonker]

Previously, something like

{
    let a := 42
    mstore(a, a)
}

would have lead to mstore(42, 42). This PR adds a bool option to the SSACFG builder, which leaves the variable and assignment intact. This can have nontrivial impact on liveness analysis and stacklayout generation. Which one turns out to be better all in all remains to be seen, hence the old behavior is not simply replaced but can be switched on and off.

[BoomchainLabs]

Lgtm

[blishko]

What's the problem with simplifying to mstore(42,42)?

[clonker]

It is not a problem per se. Sometimes though it can be useful to track constants through variables instead of (and that's the status quo for the SSA CFG Codegen) using them as push constants exclusively, as that can have a positive effect on the bytecode size. Especially for large constants. A PUSH32 needs one byte for the instruction and 32 bytes for the value. SWAP and DUP are one byte each.
While not the root cause, we currently do have some problems with codebloat in the ssa cfg codegen, so this may mitigate it a bit.
And it might make results behave a little differently when it comes to greedy algorithms for stacklayout scheduling, ie, the project that the Madrid collab is working on. In any case at this point it is something to try and see how well it fares.