refactor(pxe): avoid expensive toTx() call when computing tx hash

[nchamo]

Summary

• Replace (await txProvingResult.toTx()).txHash with await Tx.computeTxHash({ data: publicInputs }) in proveTx, and remove the corresponding TODO.

Why this is safe and worth doing

TxProvingResult.toTx() passes this.publicInputs directly as data to Tx.create (proven_tx.ts:28-29). Tx.create computes the hash via Tx.computeTxHash(fields) (tx.ts:159), which only reads fields.data (tx.ts:151-155). So calling Tx.computeTxHash({ data: publicInputs }) produces the same hash while skipping work that doesn't contribute to it:

• Walking the nested PrivateCallExecutionResult tree to collect and sort contract class logs
• Constructing a full Tx (with proof, logs, calldata) only to read one field and discard the rest

A short version of this rationale lives inline in pxe.ts.

[dbanks12]

@nchamo I think this is a bit leaky. PXE needs to know a bit about how tx hashes get computed (use public inputs, pass in via the data field).

Maybe we could add a simple helper to the TxProvingRequest instead?

[dbanks12]

That being said, this PR is a very small but meaningful change. So if my suggestion proves to be a rabbit hole, let's just merge this

[nchamo]

Like the suggestion @dbanks12 , makes a lot of sense

[AztecBot]

✅ Successfully backported to backport-to-v4-next-staging #23142.