fix(agc): transition resident attachment subresources - #6934

What

Immediate rendering to a resident mip or array-layer view transitions mip 0/layer 0 instead of the attachment subresource. APS5_SYNC_DRAWS=1 and target dumping can therefore issue VUID-vkCmdBeginRendering-pRenderingInfo-09592: the rendered subresource remains GENERAL while the attachment declares COLOR_ATTACHMENT_OPTIMAL.

Pass the selected mip and array-layer range to both immediate resident-image barriers. A 3D image still uses its single Vulkan array layer; its selected depth slices belong to that mip. Transient-image callers keep their existing default range. Only Draw.cpp changes, four lines replaced.

Tested

Current-main-only head 1b70ce290f on ea4cc112a0. Arch Linux 7.2.9-arch1-1, GCC 16.2.1 Release, Ryzen 9 7950X; RX 6650 XT / RADV Mesa 26.2.4-arch1.1 and lavapipe Mesa 26.2.4 / LLVM 23.1.1. Khronos validation layer 1.4.363, standard core checks; no explicit synchronization/GPU-assisted validation. Vulkan required, shader disk cache disabled, APS5_SYNC_DRAWS=1.

• Full libs and the three existing execution test binaries build. agc_driver_array_layer_target, agc_driver_packed_mip_tail_target, agc_driver_layered_volume_draw: 3/3 pass on RADV and 3/3 on lavapipe, zero skips and zero validation errors/VUIDs. The volume case verifies that array-layer selection is not confused with 3D depth slices; its mesh subcase validates shaders only.
• Production-only negative: replace only Draw.cpp with exact current-main source, keep the fixtures unchanged and rebuild. All three pixel tests still pass, but the array-target fixture reports two 09592 validation errors per driver, both on array layer 1 remaining in GENERAL. Restore/rebuild the fix: 3/3 pass with zero errors on both drivers again. The functional tests alone do not detect the API error; the validation logs do.
• Native Override 2 PPSA01907 01.007.000 on integration c01f24339a1b (main b260d8d792 plus fix(agc): allow chunked imports as vertex and index buffers #6913/fix(agc): restrict sampled storage texture view usage #862/fix(agc): include the host stage in the CMASK clear's host-access barriers #6781 and diagnostic readbacks): the preceding target-dump run reports ten capped instances of the same VUID on mip 1. With this fix, the repeated target-dump run exits 0 after 97.64 s with zero validation errors/VUIDs and restored settings. This integration is not isolated current-main title validation. Very dark 3D rendering remains unchanged; no visual or FPS improvement is claimed. Bring-up and exact source.
• External composition evidence from @Amoenus: feat(agc): back extra guest mip levels with storage images #7085 retains this exact 1b70ce290f3a2ef4d751c1f228540a589d3c1ff9 commit and declares it as a prerequisite. Their composition b605c6887 passes extra-mip, image-level, array-target and packed-tail fixtures 3/3 on RTX 5080 / driver 32.0.16.1062 and Radeon Graphics / driver 32.0.21036.18, Windows 11 Enterprise 26200, with Vulkan required, shader cache disabled, APS5_SYNC_DRAWS=1 and Khronos core 1.4.363: zero validation errors/warnings for those four groups. They also report Linux Mesa 25.2.8 llvmpipe 3/3 passes without explicitly enabled core validation. The layered-volume fixture passes 3/3 but emits ten WARNING-VkImageSubresourceRange-layerCount-compatibility warnings per physical run, including transfer barriers; these are not attributed solely to this change. These are attributed composition results, not standalone tests of this unchanged PR head.
• External Apple-port evidence from @jotitadev: Apple M2 Max, macOS 27.2, x86-64 under Rosetta, MoltenVK 1.4.2 and Khronos validation 1.4.363, shader cache disabled and APS5_SYNC_DRAWS=1. On macos-port at b7eb6b6b plus ImportWatched, they applied this PR with the layer-range arguments adapted to one layer because that port lacks ColorTarget::layers. The array-target fixture passes pixel checks on both trees, but two layer-1 layout errors (VUID-vkCmdDraw-None-09600) on the unchanged port become zero with the adapted change. Packed-tail has no layout errors on either tree, so this does not validate the mip fix. Both trees retain the same twelve unrelated multisample-storage image-creation errors (00968). Their broader suite reports 14/433 failures, including a cross-workgroup case reproduced 5/5 on the unchanged port. This is attributed adapted-port evidence, not a clean-main macOS run or zero-error full suite.
• Conventions against origin/main and whitespace pass. Required overlap check finds no confirmed overlap; its 22 conflicted shared-file heads were inspected at this helper and the resident barrier calls, with no selected-subresource implementation found. This is a scoped review, not a claim those whole branches combine cleanly.

Sanitized report, negative/restored validation logs and patch, assets prefixed resident-subresource or subresource-isolated. No new fixture, driver dependency or diagnostic code is included here. Standalone Windows/NVIDIA/macOS and full hosted builds have not been run by us; the external composition coverage above does not replace them. Full platform jobs require maintainer dispatch.

Checklist

• Based on current main; no other open PR implements the same functions
• One topic per PR; follow-ups go in a new PR
• No comments in code except technical debt
• Unimplemented paths throw (NotImplemented_nid_no_patch); silent stubs are listed under Silent stubs in their library's technical debt file
• No notes, investigation .md files or images added to the repository (attach them to this PR)
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Unblocks: PPSA01907 Override 2: Super Mech League (PPSA01907) #6933 (synchronous-path layout error only; dark rendering remains unresolved)
• GPUs tested: AMD RX 6650 XT / RADV Mesa 26.2.4-arch1.1, Arch Linux 7.2.9; lavapipe Mesa 26.2.4 / LLVM 23.1.1 on the same host
• AI-assisted: yes