JSON.stringify: transfer completed output into managed storage

Parent: #473

JSON.stringify builds its canonical output in the precise heap temporary allocator, then Value.strAlloc copies that completed byte sequence into another canonical allocation before managed string storage can consume or transcode it. The ownership-aware string path already accepts allocator-owned bytes and handles canonical/flat conversion plus cleanup.

Executed exact-parent evidence on real zig-gc, stripped ReleaseFast, frozen representative_json rows (2,200 jobs x 90 round trips):

• checksum: 324952086 on parent and candidate, base and variant
• backing allocations: 1,793,004 -> 1,595,004 on both rows (-198,000, exactly one per stringify)
• backing requested bytes: 597,332,140 -> 582,348,852 base (-14,983,288)
• backing requested bytes: 597,282,868 -> 582,324,324 variant (-14,958,544)
• GC cells, shape transitions, and property-lock counts unchanged
• one diagnostic CPU pass: base 588.876 ms -> 580.944 ms; variant 573.591 ms -> 574.729 ms (timing publication remains blocked until a clean host)

The proposed fix transfers the completed ArrayList slice through Value.strOwned using the allocator that owns it. This is not a JSON fast-path or external implementation; it removes a redundant whole-result copy while preserving the existing managed-string ownership and physical-storage conversion path.

Definition of done:

• allocation-failure and string-storage unit coverage passes
• pinned Test262 JSON subtree and VM agreement remain clean
• no-GIL TSan stringify witness remains clean
• exact-parent frozen checksums and counters are recorded
• clean-host seven-sample publication under #473 before closing

Landed and pushed to main as 14cbffb9.\n\nQualification:\n\n- focused JSON/OOM/Unicode/GC tests: 26 passed\n- Test262 test/built-ins/JSON: 165 passed, 0 failed\n- forced VM agreement: 165 agree, 0 diverge\n- no-GIL TSan witness: 2 passed\n- frozen full checksum: 324952086 on parent/candidate, base/variant\n- exact reduction: 198,000 backing allocations on each row\n- requested bytes: -14,983,288 base; -14,958,544 variant\n\nLeaving this open only for the clean-host seven-sample publication required by #473. The current unreapable host process makes wall-time evidence non-publishable; the process-local allocation evidence is exact and uncontaminated.