Investigate why reth doesn't work correctly when benched without --wait-time

[mediocregopher]

Describe the feature

When benching using reth-bench or reth-bench-compare, we always include a --wait-time argument of 400ms or more in order to obtain good results. If we don't do this we end up seeing blocks pile up in memory, not getting flushed in time.

We should investigate why this happens and implement a solution. Possible causes include:

• Async persistence simply takes too long
• Saving execution cache between blocks takes a significant amount of time (this has been a cause of this sort of issue in the past)
• TrieInput pre-computation takes too long

If the problem turns out to be a bug then that bug can be fixed and this issue closed. If the problem is more fundamental then we can use the parent tracking issue to organize more work.

Additional context

No response

[gakonst]

As a first step, let's find the breaking point --wait-time? My guess is that 400ms might be too long, and in general 0 wait tiem is unrealistic, so if we work well at say 50ms or 100ms then it might be fine?