Introduce Persistent Record for libafl-fuzz

[R9295]

CI should work now too!

[domenukk]

Just a thoguth, I don't think the last N always makes sense / you'd want all since restart (which, sadly, may be target dependent)

[R9295]

@domenukk AFL++ does this based on a provided value. I think it's fine.

[domenukk]

If we put this into the state, we can actually use this for InProcess fuzzing, too!
Wait, it doesn't need to be in the state since we don't need inputs prior to restarts, right?

Anyway, we definitely want this in the main lib, too! :)

[domenukk]

CI is unhappy:

thread 'main' panicked at src/main.rs:99:21:
Failed to run launcher: KeyNotFound("libafl::schedulers::minimizer::TopRatedsMetadata not found",    0: libafl_bolts::Error::key_not_found
             at /home/runner/work/LibAFL/LibAFL/libafl_bolts/src/lib.rs:350:40
...
      <libafl::schedulers::testcase_score::CorpusWeightTestcaseScore<S> as libafl::schedulers::testcase_score::TestcaseScore<S>>::compute
             at /home/runner/work/LibAFL/LibAFL/libafl/src/schedulers/testcase_score.rs:324:28
      libafl::schedulers::weighted::WeightedScheduler<C,F,O,S>::create_alias_table
             at /home/runner/work/LibAFL/LibAFL/libafl/src/schedulers/weighted.rs:174:26
...

[domenukk]

Oh wait that's in main lol

[domenukk]

It's so simple, really cool!
We should totally move this to the main lib though.

[R9295]

Regarding the inputs prior to restarts, I'll have to verify.

[domenukk]

I think it's fine, we don't care about inputs before restarts. The idea is to have a (semi-perfect) rerun of all inputs from the last restart to the crash.
For InProcess it would be good to have an option not to pop any inputs