ATTACH OOM fault-injection indices shift vs stock

The `attachmalloc` suite pins allocation-sweep indices. Extra DoltLite open-time work (vec1, `dolt_tests` / `dolt_test_run`, `dolt_ignore`, parent-dir Access probes, checkpoint discovery) adds allocations, so stock's Nth injected OOM is not ours.

This is the same representation problem as closed #1859 (`backup_malloc` per-index pins). It is **not** a missing page-image API; those five gates previously cited #1839 by mistake.

Work, if any: stop pinning per-index `attachmalloc-1.transient.N` entries (whole-suite disposition, or a harness that matches on the error rather than the allocation ordinal). Until then the five `unsupported` gates cite this.

Start with the attachmalloc suite and its allocation-sweep pins, then compare the five unsupported gates with the representation discussion in #1859. Determine whether the suite should stop using per-index attachmalloc-1.transient.N entries or use error-based matching; done means the stock-versus-DoltLite allocation shift no longer makes these gates misleading.

Written by the indexing model from the issue text.