Fine-grained Pending state

[NingLin-P]

The Pending state is a list that is used to keep track of all the (transaction, status, receipt) in the block during block execution, and then in on_finalize these data are moved to CurrentBlock/CurrentReceipts/CurrentTransactionStatuses for RPC service usage.

Because the Pending state is a StorageValue<_, Vec<...>>, which presents as a single value in the trie, appending items to the list means read,update,write to the same value, which gets slower and slower as the list grows during block execution, and depend on the number of tx in the block it can result in 3-20x of slow (see below).

This PR fixes the issue by refactoring the Pending state, to replace StorageValue with StorageMap so each tx is present as a single value in the trie, although it results in more db delete in on_finalize but in practice, it is much faster than updating a big single list.

Below is a table about the block execution time of different Pending states where N represents the total number of Ethereum::transact in the block, each tx transfers funds from the same Alice account to a new different account. The test is running on a MacBook Pro, Apple M1 Pro, 10 Core(s)

Tx in block	N = 1000	N = 2000	N = 4000	N = 5000	N = 6000	N = 8000
Current Pending state	690ms	2_498ms	9_493ms	14_921ms	21_383ms	37_426ms
No Pending state	194ms	382ms	780ms	1_047ms	1_187ms	1_596ms
Fine-grained Pending state	221ms	449ms	869ms	1_103ms	1_329ms	1_772ms

[conr2d]

Interesting results. I thought the Polkadot SDK was optimized enough for this case, but the benchmark you shared suggests otherwise. Could you share the script you used for benchmarking?