Sharded TxnLookup index: non-existing hash (or old hash) may random-touch all files

[AskAlexSharov]

Seems txn lookup by hash start slowly suffer from too many block snapshots - because we don't have cover-index (also reduced snap size to 100K).
Each lookup of old txn causing - random touch of all files tx.seg files (by false-positives of recsplit index).
Bor-mainnet has 286 tx.seg files (+ 286 tx.idx files).

Vmtouch output:
Uploading vmtouch_erigon.txt…

Reporter: https://discord.com/channels/687972960811745322/789381905958699018/1209824609206149140

Solutions

• Make false-positives cheaper: change format of tx.seg from list_of(tx_hash_1st_byte+txn_rlp) to list_of(tx_hash_1st_byte)+list_of(txn_rlp) - means all first bytes will have data-locality and all tx_lookup operations will come to that part of file. First byte of hash we have to reduce false-positives of recsplit.

• Different sharding key of tx.seg: block_num is not set in stone.

• Different indices: btree (too much keys), bloom (too much keys).