feat: use non atomic tracker for snapshot reverts

[mattsse]

closes probably #6355
possibly related #6450

Some background:

due to ds-test legacy behaviour failures are stored in a state variable.

Prior to #4974

Fuzzing was having soundness issues with snapshot failures. This is because snapshot failures were not being persisted throughout fuzz runs, and when the Fuzz Backend Wrapper we use was dropped, this information was lost, leading to inconsistent test PASS states.

because we were using a Cow of the Backend so regular fuzz runs don't need to clone it per fuzz call, however as #4974 highlighted this is unsound for snapshots because we lose state that contains the the state variable that tracks errors

#4974 tried to fix this by persisting snapshot failures in an AtomicBool. But this is also unsound because this is susceptible to race conditions because this bool is now shared by all tests, even if we clone the Backend (Arc).

The fix is to use a simple boolean again and instead add the has_snapshot_failure to the RawCallResult that we then can also check when checking for failures.

That being said, the Executor + ContractRunner, especially the Fuzz variants desperately need a cleanup.

[Evalir]

man what a find. I should've noticed the problem with the atomic back then, which I probably would've caught now...

[mattsse]

I didn't think about it either, and the ds-test failed var limitations are easy to miss.
The executor suffers massively from feature creep.
I should have cleaned that up a long time ago...

[DaniPopes]

typo