fix: race condition when reading and writing to inflight queue

[EnriqueL8]

The policy loop frequently read and writes from the inflight queue, more specifically:

• It will load new transactions into the queue when it has space in updateInFlightSet and remove the ones that have completed.
• Remove transactions from the queue when it gets enough confirmations for that transaction

When loading new transactions, the code will take a copy and override the inflight queue with an empty array. If another thread was reading from such array, most often iterating to find a specific inflight tx it wouldn't find it.

This PR alongside with adding extra logging to help diagnose this sort of issues, adds a read/write mutex for the inflight queue and a mutex for each pendingState to be thread safe. Creates read locks when iterating and reading from the queue, write lock in a single place where we remove or load new transactions in queue and a mutex on the pendingState when we read or modify it.

Draft: as running another performance test for the latest commit

[Chengxuan]

Minor comments on the loggings. The use of read locks looks good.

[Chengxuan]

LGTM