performance: keep trie prefetch during validation phase

[brilliant-lx]

Description

Recently, we saw some nodes spent lots of time on block validation phase, could even longer than 6 seconds, which leads to slash of some validators. We found out it is caused by some giant transactions, like XenCrypto. Take this transaction for example: https://bscscan.com/tx/0x5d948f50b9f1245fc4c9feaff71e02c607468d9540c33606253fa89998e9b7d8, which consumed ~40M gas in a single transaction.
It is quite unfriendly to the current trie prefetch logic, which is terminated right after entering validation phase, since getTrie() will stop the trie prefetch.
This PR tries to keep trie prefetch until validation phase is completed, which will call StopPrefetcher() at the end of StateIntermediateRoot().

Test Result:

For these long time validation phase, the validation time could be reduced by more than 80%, which is a huge improve.

//= latest develop branch: import the trouble  block take ~12s
t=2023-10-30T17:11:19+0000 lvl=info msg="Imported new chain segment"            number=33,060,898 hash=0x59e91888d0b6a5aef78e961ef5eb7545eed3aad3409cc26a38625639f4225c89 miner=0xb218C5D6aF1F979aC42BC68d98A5A0D796C6aB01 blocks=1          txs=115           mgas=24.430  elapsed=11.968s      mgasps=2.041   dirty="86.95 MiB"

// develop with this patch: import the same  block take ~2.1s
t=2023-10-30T17:11:09+0000 lvl=info msg="Imported new chain segment"            number=33,060,898 hash=0x59e91888d0b6a5aef78e961ef5eb7545eed3aad3409cc26a38625639f4225c89 miner=0xb218C5D6aF1F979aC42BC68d98A5A0D796C6aB01 blocks=1          txs=115           mgas=24.430  elapsed=2.137s       mgasps=11.431  dirty="278.60 MiB"

Metrics of the test:

Rationale

NA

Example

NA

Changes

NA

[RenRick]

LGTM

[brilliant-lx]

reference: #952