Rollback #6713 due to possibility of dropping transactions received right before an epoch boundary.

[nikurt]

No description provided.

[jakmeier]

Feels like too little context. Could you write down somewhere (GH issue, comment, this PR description, or previous PR) why the current solution does not work? I can sort of guess it from the title but as someone unfamiliar with this code area, I can't say I understand the details. And hence I cannot review it with good faith.

[nikurt]

@jakmeier That's a fair question. I have a more verbose issue description in https://docs.google.com/document/d/1tZvXyaQeFSMUY-IyFtU_vjkkMiKco1_nLRtGzzBERMo/edit#bookmark=id.k2s0x3ipu5bt

The root cause of the issue is that a transaction is forwarded either to:

• chunk producers of the next 4 blocks in the current epoch
• chunk producers of the next 4 blocks in the next epoch

The current solution doesn't work if:
a) a node forwards transaction to the next epoch chunk producers, but
b) those chunk producers are not validators in the current epoch

because the forwarded transactions will simply get dropped.

The previous solution worked well enough in practice and rolling back to it should be fine.