Rework banking_stage retry logic to retry high priority transactions more often

[tao-stones]

Problem

High priority transactions that are not processed immediately (eg due to account_in_use) are not retried until entire buffer has been iterated.

A specific example found during tds bench test: a batch of 128 prioritized transaction that write to same account were received and put on to top of banking_stage buffer; the remaining buffer was full of zero-prioritized transactions. the top 128 paid TXs were picked up but only one was processed, the remaining 127 TXs were re-buffer, but they were not retried until after entire buffer.

Proposed Solution

Before the scheduler, can try to mix TXs in retry queue and what's left in MinMaxHeap for each chunk process.

[tao-stones]

Multi Iterator implemented in https://github.com/solana-labs/solana-ghsa-qrp2-62qh-jgfc/pull/6 addressed this.

#28577 proposed another possibility.

Close this issue.