node/consensus: priority lock for CE business to preempt tx influx

[jchappelow]

This adds node/consensus.PriorityLock that is like a normal mutex but:

• Lock guarantees FIFO when acquired, no issues with sleeping goroutines in the sync.Mutex breaking ordering
• PriorityLock jumps the line, getting the lock ahead of others that used Lock

This is intended in the very specific use case of CE where the mempoolMtx was locked in:

• commit: block commit, priority
• prepareBlockProposal: leader preparing block proposal, priority
• QueueTx: ingest new tx from either p2p announce or RPC broadcast, NOT priority

All of the "priority" cases are sequential, never concurrent with each other. Only the QueueTx spot can be concurrent and at any time, as it is triggered from external sources.

This eliminate the one bad spot of mutex contention revealed by running (for a brief period) with --profile-mode=mutex while stress tool is flooding the network.

[jchappelow]

I need to doc and clean up this PR.