feat: return Database::error

[Wodann]

First step to resolve #309

I wanted to get this under your eyes, to see what you think. This only aims to bubble up the actual FatalExternalErrors that should not result in any gas cost.

[rakita]

This is not optimal as there is a lot of boilerplate code and if you are in 1000 call depth you would copy/move the DB error 1000 times. That is why I have only the simple stupid enum Return so I am moving only one byte and checking only one byte in the interpreter hot loop.

and is Debug no_std, I want revm to be able to be compiled for wasm?

[Wodann]

To make better usage of the memory, we can include the Reason in the Result::Ok to share the memory. That would allow us to remove it from the interpreter.

Also, the compiler should be smart enough to move the memory in one go for the Reason instead of moving it a 1000x to. If I'm not mistaken, me flagging it as Infallible for some paths should even result in complete optimisation of the code path. So for a Database::Error = Infallible, it should have no performance overhead. Benchmarks would be required to confirm that, though

[Wodann]

and is Debug no_std, I want revm to be able to be compiled for wasm?

I can include core::fmt::Debug instead

[rakita]

Either way, Reason change should be localized, and this PR has a lot of unnecessary changes to the code.

[Wodann]

Which changes would you say are unnecessary? Would you like to see how an implementation that uses Return::FatalExternalError with an if branch holds up against this implementation? Is there a benchmark I can use to compare?

[rakita]

I never intended to remove Return but just to cast it to something more reasonable to external call (as in from transact).

In the interpreter, Return is not made idiomatic and the reason why I didn't use ? is there were reports of it being slower and LLVM was not optimizing it: https://agourlay.github.io/rust-performance-retrospective-part3/

Other than that you are adding more code to every instruction to fit Result pattern, it is unnecessary.

[Wodann]

I made a minimal implementation, as (I think) you envisioned. This should be a lot less intrusive. It also resolved a bunch of in-code TODOs

[rakita]

Change introduced here: #334