chore: bump Noir to test noir-lang/noir#13466

[asterite]

Checking whether aztec-packages CI passes with noir-lang/noir#13466 (disabled-branch array_get fallback biased to the type-matching field's flat slot offset, generalizing compute_offset beyond single-slot-field arrays).

This is the retry of the change previously attempted in #21061 / noir-lang/noir#11744, which failed CI here: that version computed the offset in item ordinals while the consumer applies it to the flattened index — misaligning the disabled-branch fallback read for any element with a multi-slot field before the accessed one. The unit mismatch is now fixed and pinned by a regression test (regression_11744) on the Noir side.

Do not merge — testing only.

🤖 Generated with Claude Code

[asterite]

Repointed the pin: noir/noir-repo is now v1.0.0-beta.26 (40d6574f, the exact commit next already pins) plus only the noir-lang/noir#13466 commits cherry-picked on top (3ad4ecf5dd4c, branch ab/acir-generalized-array-offset-beta26).

The previous push pointed at a noir-master-based head, whose unrelated dependency bumps (toml 4→5, tslog 5.1, …) invalidate avm-transpiler/Cargo.lock / yarn-project/yarn.lock without a full bump_noir_compiler.sh run — the earlier CI failure is most likely that, not the compiler change. With the beta.26 base the only delta CI sees is the compiler change itself, so this run's result is the meaningful one.