feat(agc): implement image tables loaded through 64-bit pointers

[Celegans12]

What

Stacked on #757; please review the last commit only.

Image and sampler descriptors loaded by s_load at runtime offsets through a 64-bit pointer still fail planning. This enumerates offsets from an independently resolvable base, preserving registration-time compilation and keeping invocation data out of shader and pipeline keys.

Consecutive descriptor reads must share one heap/address handle and an affine offset. Alignment, signed immediates and bounded keys are preserved; invalid, unmapped or overflowing selected entries fault. Each table records its consumed descriptor spans, read in runs like #757's buffer tables. Overlapping buffer or storage-image writes poison that table, while gaps and unrelated tables remain usable.

Pointer columns share #757's column rules: reads with the same access shape share one column, whose key bound is the widest of their keys and whose stored plan keeps the pointer flag and bound. Records that hold a pointer to the T#, and a T# split over two loads of its record, are not planned here; #4475 (stacked on this) adds both. #5466 needs a table-aware rule when it meets this, which its author described there. Persisted invocation ranges advance disk format 32 to 33; runtime ABI 12 is unchanged.

Tested

Based on main 066b58f with #757. Windows 11 (MinGW-w64 GCC 15.2.0) and Linux: configure -DBUILD_TESTING=ON, full build, --target libs, ctest.

• Windows, NVIDIA: 1188/1189 tests pass; log_macros also fails on main here.
• Linux (upstream's job on WSL, lavapipe), merged with main 84b01f9: 1171/1173; agc_driver_bc_block_copy and agc_driver_depth_compare_copy also fail on main there.

The pointer cases in agc_driver_image_table (masked keys, null and invalid entries, key cap, unmapped entries, overlapping storage writes, two-table isolation), agc_shader_memory and agc_shader_disk_cache (a pointer table's plan round trips) pass on both; they fail to build without this commit. Windows ran counted reuse with write watch; WSL skips it and the two subgroup-32 per-wave cases.

Not tested: graphics-stage pointer tables, native RADV, console behaviour.

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: feat(agc): implement image and sampler tables indexed at runtime offsets #757
• Unblocks: PPSA26344 Ghost of Yōtei (PPSA26344) #675
• GPUs tested: NVIDIA GeForce RTX 3080 Laptop GPU, driver 610.62, Windows 11; llvmpipe (Mesa 25.2.8), WSL Ubuntu 24.04
• AI-assisted: yes

[Celegans12]

The Linux failure here (agc_driver_flat_store timing out at 120 s) is the llvmpipe slowdown from #757 that I reported in this comment. The fix is in progress on #757, and this branch will be restacked onto it.

[Celegans12]

I've restacked this on #757's new head d9e47d1 (rebased onto main 066b58f), so the head is now 34525fd. Its own diff hasn't changed, and the disk format is still 33. On this head, Windows NVIDIA passes 1,188 of 1,189 tests, and upstream's Linux job merged with main 84b01f9 passes 1,171 of 1,173; the failures there (log_macros, agc_driver_bc_block_copy, agc_driver_depth_compare_copy) are the same ones main has.

[adriadam10]

Stacked on #757, reviewed together: see the review there (#758 head 34525fd merged with #757 onto main 84b01f9, Linux RADV 572/573, the one failure also fails on main). The pointer cases in agc_driver_image_table, agc_shader_memory and agc_shader_disk_cache pass. Approving; it goes in after #757, each with the full Build dispatched on its final head.