Closes neo-project#1991.

Fix the comment neo-project#2818 (comment).

Fix Vitor's comments.

Fix Owen's feedback.

Consider the case when transaction B has the hash of transaction A
in the Conflicts attribute and transaction B is on-chain. Let the
transaction C has the hash of transaction A in the Conflicts attribute.
Then transaction C still should be able to pass verification and should
be accepted to the subsequent block, because transaction A isn't on chain
and transaction A will never be accepted.

Thanks to Owen for testing, this case is described at the
neo-project#2818 (review).
The expected behaviour in the described case is that TXID-D and TXID-E will
be successfully accepted to the mempool and to chain, because the conflicting
TXID-A is not on chain (and it's OK that the hash of TXID-A is on-chain as
the conflicting hash).

Testcase provided by Vitor in neo-project#2818 (comment).

Transaction that conflicts with mempooled transactions have to pay
larger network fee than the sum of all conflicting transactions in
the pool that have the same sender as the newcomer.

Port the nspcc-dev/neo-go#3031.