fix: slice undersized crs in browser (backport #24894)

[AztecBot]

Manual backport of #24894 to v5-next, requested in Slack (release 5.1.1/5.2.0 scoping — the fix is blocking browser work).

Clean cherry-pick of cbd096d (-x), no conflicts. One file, +5/-1: when the browser cache holds a larger CRS, slice it to exactly numPoints * 64 bytes so the WASM compressed/uncompressed auto-detection does not throw invalid points_buf size. On v5-next the file lives at barretenberg/ts/src/crs/browser/cached_net_crs.ts (it moved to barretenberg/ts/bb.js/src/... on next). The resulting blob is byte-identical to the one merged on next (2de1743).

Why the automatic backport never ran

#24894 did carry backport-to-v5-next, and backport.yml fired on both labeled (09:19) and closed (11:44) on 2026-07-22, but each run skipped the Backport/port PR job because the label gate reported failure:

• dfb6942 (2026-07-16) set the gate to agilepathway/label-checker with prefix_mode: true and any_of: backport-to-,port-to-next. That action only supports a single prefix. The merge-event run log shows it verbatim:

  ##[error] Currently the label checker only supports checking with one prefix, not multiple.
  Label detection status: failure
  

  So steps.check.outputs.label_check was failure, the job's state output was failure, and the backport job's if: ... needs.label_checker.outputs.state == 'success' was never satisfied → Backport/port PR = skipped.

• The failure was silent: allow_failure: true kept the Check labels job green (so the whole run concluded success), and both the "❌ Failed to cherry-pick" PR comment and the #backports Slack alert live inside the skipped backport job, so neither fired. Nothing anywhere said the backport had been dropped.

• The gate was rewritten as a direct jq prefix check in 010569d (feat(ci): auto forward-port v5 merge-train PRs to next #24848). That fix reached next in merge commit cc4f066, whose first parent is b5cd35c — this PR's own merge commit. In other words the fix is literally the next commit on next after #24894, so #24894 was the last PR to merge under the broken gate. (Commit author/committer timestamps are misleading here: they record merge-queue entry, not landing order — ancestry does.)

Affected window

PRs labeled backport-to-* / port-to-next that merged between dfb6942 (2026-07-16) and cc4f066 (2026-07-22) had their backports silently dropped. Checking the backport-to-v5-next PRs that merged in that window against v5-next:

PR	On v5-next?
#24894	❌ missing — fixed by this PR
#24830 fix(ci): move nightly bench inclusion sweep to 06:00 UTC	❌ still missing (CI schedule only, low impact)
#24843	✅ landed later via #24961
#24827	✅ present (71f8e3d, arrived by another route)
#24804, #24907	✅ backported by hand (#24808, #24945)

The automation itself is fixed going forward — backport-to-v5-next-staging is currently drained and in sync with v5-next.

---

Created by claudebox · group: slackbot · Slack thread