fix: Reduce view scope in batch_builder.get_batch_inputs

[sergerad]

Summary

We have noticed that the sequencer reports 2 live state snapshots (snapshots.live on the store's write_block span) far more often than full nodes do, even though full nodes serve more read traffic.

The likely cause is BatchJob::get_batch_inputs. It bound a StateView for the whole function, so the snapshot stayed pinned while the fee collection transaction was proven locally and validated by the validator. Neither step reads the state. When that work overlaps a block commit, the previous snapshot generation is still pinned at publish time.

This PR releases the view as soon as the last state read is done.

Changes

• Scope the view with State::with_view so it only covers the store reads and the fee collection transaction execution.
• Split get_batch_inputs into three helpers:
  • read_batch_inputs: reads the note inclusion proofs, partial blockchain and reference block header.
  • execute_fee_collection: reads the protocol config and genesis commitment, then executes the fee collection transaction.
  • prove_fee_collection: proves the transaction and validates it with the validator. Runs after the view is released.
• Add a private BatchInputs struct to carry the store reads between the helpers.

No behavior change is intended beyond the shorter snapshot lifetime.

Testing

Existing batch_builder tests pass. They do not cover the fee collection path.

Changelog

changelog = "none"
reason    = "Internal change only."

[kkovaacs]

Nice!