fix(pxe): run parallel operations to completion on failure

[AztecBot]

Port of #25208 onto merge-train/fairies.

Original description:

The PXE stores intermediate simulation state (contract classes, notes, tagging indexes, etc.) via a set of KV stores. To avoid the tx being polluted with intermediate state on failure, most of these stores accept 'staged' entries that only get persisted if the tx succeeds.

The PXE runs many things in parallel to speed things up, using Promise.all. This is problematic because as soon as one operation fails, we discard the tx and drop its staged writes — but siblings continue running and may recreate a staged entry after we drop it, wedging reorg handling and leaking memory.

This PR adds an allToCompletion helper to foundation that runs a collection of tasks fully (mirroring Promise.allSettled semantics) but rejects with the first error once every task has settled, and replaces every Promise.all in PXE with it.

Cherry-pick of 8ac210655c2ea74f708caea05cab55f15575ae65; applied cleanly on top of merge-train/fairies.

---

Created by claudebox · group: slackbot · requested by Nico Chamo · Slack thread