Parent: #498 Related: #1074, #1076, #1077, #491
The live inventory classifies packed-array push as legacy_pattern_kernel / legacy_narrow. The VM recognizes a counted loop containing one intrinsic push, decodes its numeric argument expression, and batches up to 64 calls into a vector push. It also recognizes one add-three-and-bitmask expression. The compiler emits a corresponding structural candidate hint. This conflicts with the no body/benchmark recognizer constraint.
Executed baseline on 3ffad54f: the standard fill-plus-sum witness passes 2/2, asserting push-kernel hits in isolated and synchronized modes while returning 36 with 316 logical steps. The dense-array witness passes 2/2, asserting both push-kernel and add-three/bitmask-specialization hits; fill(8,2,1) returns 806. Own/prototype overrides retain their observed results.
Acceptance:
- Remove the packed-push candidate hint, body recognizer/executor, numeric-expression decoder/specialization, push-only types/constants, plan tag, and hit metrics. Retain general intrinsic-push and indexed opcode paths.
- Ordinary calls preserve result 36 / 316 steps in both modes and the expression witness 806, plus argument evaluation, method/property lookup, receiver effects, overrides, partial effects, and live bindings.
- Actual no-GIL workers, explicit moving-GC roots, source hooks, budget/stop checkpoints, and allocation failures remain covered.
- Focused normal/ReleaseSafe/unsuppressed TSan checks and exact before/after affected test262 and tree/required-VM comparisons show no new failures.
- Regenerate the live inventory/docs/manifest, retain shared helpers needed by other live families, and preserve historical evidence with remaining legacy work explicit.
This removes body-specific work substitution; it does not establish current throughput or complete #498.