perf(core-transaction-pool): use tree memory structure to sort by fee and by sender nonce

[air1one]

Summary

Plus other perf improvements.

Now inserting txs is slower (expect around 10k per second, still okay).

But fetching is way faster, which is important when getting transactions for forging, considering default tx pool limit is 100k.

Checklist

• Documentation (if necessary)
• Tests (if necessary)
• Ready to be merged