ingest/backfill.go: patch to always end backfill at checkpoint aligned ledger

[cjonas9]

What

Ensures the final right boundary of the fill/final filled ledger is always aligned to a checkpoint.

Why

Previously, despite checkpoint-aligning the right frontfilling bound in runFrontfill(), this bound would be re-set to the most recent ledger in the datastore in runFrontfillChunks(). This caused noisy logs as captive core would attempt to pull a future checkpoint that hadn't yet been published, that would be resolved when that checkpoint was published. Also, this caused a rare, nondeterministic bug, where a UNIQUE constraint violated error could occur under the following conditions:

• backwards backfill was skipped (causing two frontfilling iterations).
• in iteration 1 of frontfilling, the unaligned datastore tip advances from the first checkpoint into the next checkpoint.
• the total amount of ledgers it advances by is < 64 ledgers.

This is unlikely in practice because, by the third bullet, it must be a small backfill (i.e. under one day of ledgers in the worst case).

Known limitations

N/A

[Copilot]

Pull request overview

This PR ensures the final frontfill/right boundary used during backfill is checkpoint-aligned, preventing frontfill from accidentally extending to an unaligned datastore tip and triggering noisy logs / rare uniqueness issues.

Changes:

• Compute and persist bounds.frontfill.Last as the datastore tip aligned to the previous checkpoint in runFrontfill().
• Stop frontfillChunks() from re-reading the latest datastore ledger and overriding the caller’s right bound.
• Update log messages and advance frontfill.First based on the aligned right bound.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.