seems to work OK. Could be sped up by local filters instead of
ingesting every tx from the block but I can do that later if
performance (mostly disk i/o) is an issue
once multiple connections are implemented, hard mode should forward
txs to blend in.