[EVM] Refactor StackSlot std::variant -> LLVM style RTTI

[akiramenai]

No description provided.

[vladimirradosavljevic]

Could you please run clang-format?

[akiramenai]

clang-format is done.

[vladimirradosavljevic]

I ran tests, and I saw some issues:

1. Use-after-free issue (explained in the comment).
2. tests/solidity/simple/loop/complex/1.sol is failing with Variable not found on stack, where previously was failing with too deep in the stack.
3. tests/solidity/simple/algorithm/floating_point_simulation/geometry/lines.sol is failing with too deep in the stack, where previously was passing.

[PavelKopyl]

As we have switched to pointer semantics, I think it's reasonable to add const qualifier to StackSlot pointer definitions where this semantically required and also removing auto. At least this will simplify the code reading.

[akiramenai]

The same tests are failing now in the base branch and the PR.
So the patch is ready for another review iteration.

[vladimirradosavljevic]

LGTM, thanks! Just a comment to reduce repeated hash lookups (we have up to 3 per call currently).

[vladimirradosavljevic]

Suggested change

	if (LiteralStorage.count(V) == 0)
	LiteralStorage[V] = std::make_unique<LiteralSlot>(V);
	return LiteralStorage[V].get();
	auto [It, Inserted] = LiteralStorage.try_emplace(V, nullptr);
	if (Inserted)
	It->second = std::make_unique<Literal>(V);
	return It->second.get();

[vladimirradosavljevic]

This also applies to methods below.

[akiramenai]

I prefer to keep the original implementation even if optimizer doesn't manage to remove the second lookup. It's shorter and better readable.

[PavelKopyl]

LGTM, thanks!

[PavelKopyl]

Please, rebase also both stackifcation branches into the main.