Refactor append_tx_effects_for_blob to avoid duplicated computation

[TomAFrench]

See #11037 (comment)

[MirandaWood]

Did you address this in #11213 or is there more that can be done to dedupe some code?
Not sure whether we could merge get_tx_effects_hash_inputand get_tx_effects_hash_input_helper and call it twice, once constrained once unconstrained, and have some if std::runtime::is_unconstrained() clauses?

[TomAFrench]

I don't know if I'd say that PR closed this issue but at some point we'll hit diminishing returns.

We can definitely wrap the constraining logic in if !std::runtime::is_unconstrained() to improve simulation speed.

The main potential optimization I can see is that it's probably worth making an unconstrained helper for this function (and potentially removing the usage of boundedVec?).

aztec-packages/noir-projects/noir-protocol-circuits/crates/rollup-lib/src/components.nr

Lines 524 to 535 in be382bc

	fn get_all_update_requests_for_tx_effects(
	all_public_data_update_requests: [PublicDataWrite; MAX_TOTAL_PUBLIC_DATA_UPDATE_REQUESTS_PER_TX],
	) -> [PublicDataWrite; MAX_TOTAL_PUBLIC_DATA_UPDATE_REQUESTS_PER_TX] {
	let mut all_update_requests: BoundedVec<PublicDataWrite, MAX_TOTAL_PUBLIC_DATA_UPDATE_REQUESTS_PER_TX> =
	BoundedVec::new();
	for update_request in all_public_data_update_requests {
	if !is_empty(update_request) {
	all_update_requests.push(update_request);
	}
	}
	all_update_requests.storage()
	}

[MirandaWood]

Yeah, that function costs a ton of gates. I'm not sure whether we still even need it (see comment // TODO remove this? The avm should be returning public data writes left aligned.), so I think that's why it's untouched. @sirasistant do you know if we can remove this function?

[sirasistant]

Yes I think we can just remove it! I think in the past we had that because we had 63 updates from the avm and then we always put the fee write in position 63. I think now everything should come left aligned, and if not, it's a bug elsewhere!

[TomAFrench]

I think we can close this now. The remaining duplicated computation is just encoding the prefixes which would be harming readability for negligible benefit.