[Roxas2712]

Native Gen2 camera controls could not install because the native FieldMoves facade lacked bike/surf state predicates. After an external free-body landing, ice, currents and forced doors also needed to return to the existing native mover. This change supplies the native state queries and forced-terrain handoff while preserving native collision, queue interpolation, completed-step ownership and reentry guards. It also corrects Apple Silicon being classified as a low-end ARM handheld, which incorrectly reduced the supplied Johto package's AUTO graphics profile.

The implementation exposes only the existing native movement states; legacy FieldMoves action/fly-picker APIs remain open. Explicit performance selections remain available. Map changes retire the ice latches.

The included runtime qualification tools and dependency patch are experimental: increasing the LuaJIT capacity to 120 allows the unchanged supplied renderer extension to compile and initialize in a separately copied LÖVE app. Cross-platform qualification and component staging are included; this source PR alone does not replace the distributed runtime.

Stacked prerequisites: #204 (persistent world/player/body phases), #207 (native collision reads), #235 (terrain queries), #236 (special handoff admission). Their overlap is resolved in this independent composition. The fork's launcher/release/HTTP integration changes are not part of this branch.

Validation is source-bound in the docs and sanitized receipts:

• Native focused suites: 30 field-state, 584 forced-handoff, 21 body-phase, 110 live-state, 549 special-handoff and 313 CountStep assertions. The unchanged supplied controls pass 92 CPU integration checks across First/Third Person.
• Fork: 343 engine suites, three SDK suites and all quick tiers pass with both the standard and experimental arm64 runtimes. Apple Silicon policy has 42 regression assertions.
• Fork-only actual Crystal GPU checks: OpenGL and Metal, First/Third Person. The normal Mart/Center door/carpet and Cherrygrove/Route 29 matrix passes 212 checks, 12 transitions, four menu checks and four fresh owned-window proofs per backend. Separate NATIVE/2X-AA size comparisons pass.
• Experimental runtime: interpreter/JIT, shared captures, GC, debug surgery with the stock trace-flush contract, bytecode round trips and actual C-API captures through 120 pass. The renderer extension installs in both cameras on both GPU backends (16 checks and two owned-window proofs per backend). A subsequent extended-runtime transition matrix requires that installation and passes 213 assertions, 12 normal transitions, four native-menu checks and four fresh window proofs per backend. Black-and-white objects remain visible in some First Person frames; full visual parity is not accepted. No private generated source is published.

The fork's native runtime runs 40209358578 and 40210398926 now pass Linux x64, Windows x64, Intel Mac and ARM Mac stock/extended capacity, C API and engine checks, plus actual Windows/Intel/ARM LOVE host/thread/FFI gates. Source-bound component staging verifies exact receipts and architectures and combines both original Mac slices. The applicable shared upstream source-inspection fixtures normalize checkout CRLF here. These fork receipts do not claim that this independent upstream composition has passed the same hosted matrix.

Actual fork package candidates from f7a8697 have identical Windows/macOS payloads, a verified fused Windows executable/DLL and a strictly signed universal Mac app (minimum macOS 11.0.0). On Mac ARM, the exact packaged engine sources and byte-identical packaged Lua library pass 213 checks, 12 normal transitions, four menu ownership checks and four fresh window proofs per OpenGL/Metal backend in both camera modes. Own QA entry points prevent opening a release/player profile; initial positions/introduction are setup. Some proofs still contain unusual objects; Metal also displays deprecated-shader warnings. The runtime receipt records sources, hashes and exact limits. Fork preview publishing integration is not included in this upstream branch.

The GPU results are from the fork, not evidence that the independent upstream composition has been GPU-tested. New Bark's fresh-save story/owner approach remains unaccepted. All-story, remaining forced-terrain GPU, full Pokemon model/effect/battle parity, physical input, long-session, Windows GPU and Intel GPU checks remain open. This upstream PR remains a draft development checkpoint. The existing fork 20261025.1 desktop assets have now been refreshed in place from merged main 0acb23b with the qualified runtime. Exact sources, asset hashes and limits are recorded in the receipt and release manifest. The original tag/source archive is retained, with the actual binary commit linked explicitly. No supplied mod source/assets, ROM data, screenshots or saves are published.

Current fork validation at 0acb23b: four native platform gates plus component staging pass in run 40211783006; actual packaged-source/universal-runtime transition tests pass 213 checks per OpenGL/Metal backend. Explicitly staged Crystal battles additionally pass 20 checks and four window proofs per backend, with both selected field cameras, visible native colored Pokemon cards, native menu/attack input, damage and win callbacks (zero provider fallbacks). All 16 current-source proofs were inspected. This is Mac ARM evidence with setup positions/party/introduction, not ordinary visible-contact, full model or gameplay parity. Black/white overworld actors, some command clipping, Metal warnings and outdoor performance spikes remain. The published-source receipt preserves those limits and the corrected species/first-move QA assumptions.