Inefficient TxLookup prune

The [prune function for tx lookups](https://github.com/torquem-ch/silkworm/blob/master/node/silkworm/stagedsync/stage_tx_lookup.cpp#L194-L217) seems to me quite inefficient.

Basically it **traverses all records** from `BlockTransactionLookup` bucket in search of linked block numbers which are below the pruning threshold. That table has hundreds of millions records.

I believe a more efficient approach is to load transaction hashes from to-be-pruned blocks and delete records by transaction hash which is the **key** of `BlockTransactionLookup` bucket