Unprovable transactions not dropped

[LHerskind]

If having a contract that emits nullifiers without consuming a note (in my case to ensure non-replayable) you can successfully execute transaction simulations that replay the payload without being able to build a valid block because it is trying a double-spend.

This is to be expected, but unexpectedly, the transaction is not dropped, meaning that the sequencer will run into an infinite loop where it tries to build an unprovable block until it times out. This make it quite impractical to properly test replays that are not caught by the application circuit itself.

[LHerskind]

This was fixed in #2069 by doing a second round of dropping txs with duplicate nullifiers after the public transactions were executed. The sequencer seemed to rely on the assumption that only private transactions could emit nullifiers (errors should arise with custom nullifiers or when consuming l1 -> l2 messages)