crash in grimoire's generalization stage is not added to objective corpus

[R9295]

i'll fix it

This may be present in other stages where inputs are modified and Executor::run_target is used. Not the case in tracing since input is unmodified but maybe in colorization?

[R9295]

Fuzzer::evaluate_input should be used instead? Maybe the reasoning was to avoid new corpus entries? but if there's new coverage, why not add it to corpus?

[k5602]

Fuzzer::evaluate_input should be used instead? Maybe the reasoning was to avoid new corpus entries? but if there's new coverage, why not add it to corpus?

I was considering Fuzzer::evaluate_input as u suggested, but realized it might be redundant since it re-runs the target. Am I correct in assuming evaluate_execution is preferred here because it leverages the observers already populated by run_target?

I’ve implemented the PR with this approach in mind, but I’d appreciate your clarification to ensure I haven't overlooked a design constraint or a wiser choice.