Don't send the entire compact_tx in a tx_notify_request

[narula]

Right now, each shard includes the entire compact_tx in the tx_notify_request message to the atomizer when it attests to a transaction's inputs.

If a transaction spans multiple shards, the compact_tx is getting sent to the atomizer n_shards times.

Instead, the shard could send a smaller (compacter?) transaction which just includes the inputs it is attesting to. Perhaps the lowest index shard could be responsible for sending the transaction's outputs.

[wadagso-gertjaap]

I've thought about this somewhat, and came up with a few thoughts:

• I think this requires adding and additional field for the index of the inputs the shard passes
• In stead of the lowest index shard, should we have the shard attesting to input 0 attach the outputs? That way, the burden of sending the outputs is distributed amongst the shards.
• We would need a field indicating the total number of (either included or omitted) inputs in each CTxNotify to ensure the atomizer can properly determine when the transaction is complete
• Should we implement some kind of variable length integer for attestation/input indices and lengths? uint32_t seems wasteful in most cases. It would further decrease the bandwidth usage.

I'd be happy to take a stab at implementing this. Seems doable.

Another optimization further down the chain is to have sentinels only send transactions to shards that need to attest something (it seems the routing being static allows for that?). Is there a reason not to do that?