Settle the attributes iterator's per-scope state once per pass

[ericproulx]

Summary

For every element of an array scope, and for every validator on it, AttributesIterator#do_each re-derived things that depend only on the scope and the request: whether the scope iterates elements, whether its params are an Array, which scope an element's index is recorded against, and the request's ParamScopeTracker (a fiber-storage read). They are now settled once per #each and handed down, and the top-level call no longer allocates its empty parent_indices. An element-iterating scope whose params are not an Array returns before iterating, since it yielded and recorded nothing anyway.

One validator's iteration overhead on a ten-element array: 5.46 → 4.64 µs.

Missing spec, added

The branch where a lateral scope (a with group inside an array) records each element's index against its nearest array ancestor had no coverage: removing it passed the whole suite, while errors then named the wrong element (array[2][b] for the first element, and two errors collapsed into one). The new spec fails a first and a last element; it passes before and after this change, and fails with that branch removed.

Benchmarks

Median of 9 interleaved subprocess rounds against master, Ruby 4.0.6, no JIT:

request	delta
JSON POST, 10 array elements × 3 attributes	+4.1%
GET with flat params (control)	+0.2% (noise)

Test plan

• Full RSpec suite passes locally.
• RuboCop clean.
• Mutation-checked: never recording indices fails 23 specs; dropping parent indices fails 2; validating an array scope that did not get an Array fails 1; ignoring the nearest array ancestor fails 1 (the new spec).
• CI green.

🤖 Generated with Claude Code

[github-actions]

Danger Report

No issues found.

View run

[ericproulx]

Closing in favour of #2936, which combines #2922–#2934 into one PR, re-benchmarked as a whole against master. The write-up here (behaviour matrix, mutation results) still describes this part of the change.