Parent: #498 Related: #1074, #1076, #1077, #1078, #1079, #491, #477
The live inventory classifies fixed-shape object allocation as legacy_pattern_kernel / legacy_narrow. The VM recognizes one exact indexed-selector/replacement-literal/checksum loop, then substitutes batched allocation, slot writes, and loop control with a kernel-specific checkpoint protocol. It is the last consumer of the array-loop plan table and candidate machinery. This conflicts with the no body/benchmark recognizer constraint.
Executed baseline on 7decd4fc: the VM witness passes 2/2, returning 1,456 with 5,944 logical steps in both synchronization modes, matching Node. Kernel-hit, literal-shape preparation, checkpoint-boundary, budget, override, and guarded-fallback assertions pass. The shared-worker witness also passes 2/2, asserting kernel hits and batch-reserve behavior across actual Thread workers.
Acceptance:
- Remove the exact allocation-loop recognizer/executor, kernel-only plan/types/counters/test metrics, array-loop candidate flag/hint, and per-instruction array-plan table; retain general literal/index/property paths and shared helpers used by live families.
- Ordinary execution preserves result 1,456 / 5,944 steps in both modes, checkpoints/budget/stop accounting, literal prototype/shape semantics, coercion, indexed/receiver effects, and partial exceptions.
- Preserve nursery child roots, restrictions, actual no-GIL workers, explicit moving-GC roots, source hooks, and allocation-failure behavior through ordinary allocation.
- Focused normal/ReleaseSafe/unsuppressed TSan and exact before/after affected test262 and tree/required-VM comparisons show no new failures.
- Regenerate inventory/docs/manifest and update semantic witnesses without retaining kernel-hit substitutes; preserve historical evidence and explicit remaining legacy labels.
General unpublished-object reserves, prepared literal descriptors used by JSON, and observable-step helpers used by bounded leaf calls are separate mechanisms and remain where independently required.
This removes body-specific work substitution; it does not establish current throughput or complete #498.