perf: remove pending queue for MultiproofManager

[yongkangc]

Describe the feature

In previous discussions, @mattsse has mentioned that we might not even need Pending Queue as we already have crossbeam which serves as the queue.

In this PR #19171 , we realised that not having this queue might be beneficial for perf, as the pending_queue was a key bottleneck (increasing limit improved performance greatly - seen on nightly metrics as well)

Therefore we should explore remove pending queue and is_full() from MultiProofManager

We do have to keep in mind possible perf regression from https://github.com/paradigmxyz/reth/pull/18727/files which introduced chunking only when queue is full

Additional context

No response

[leniram159]

would love to clean it up

[yongkangc]

would love to clean it up

hey @leniram159 thanks for the initiative, however this requires benchmarks and something we wanna do internally. next time!