fix(agc): read the DCC keys of sampled textures after the GPU writes recorded before them

[Nesyth]

What

A sampled texture's DCC keys were read straight from guest memory. That read is textureClearKeys, which sits behind TextureClearKeys, ProvedClearKeys and the sampled path of cachedTexture. Recorded work that has not run yet can still write those keys: a title's clear or decompress kernel, or a copy or fill over the key range. A texture sampled later in the same batch then decided its content (clear-code fill or "read as stored") from the keys as they were before that work.

• textureClearKeys now reads the keys through currentDccKeys, as render targets already do through CurrentDccKeys. Under the GPU lock, a key range with a recorded writer is answered in one of two ways:
  • if the driver made that store itself, from the pending store's keys (PendingStoreKeys);
  • otherwise, the read first waits for the writer (Recorder::SyncThrough).
• Without the GPU lock, or with no recorded writer over the range, the keys are read as before.
• A sampled texture over keys that recorded work writes can therefore now wait for that work under the GPU lock, where it used to read the memory at once.
• Test: pendingKeyStoreTests records a vkCmdFillBuffer that writes 0001 clear keys (0x40) over a texture's key range, and notes it as a pending write. TextureClearKeys must then answer Clear0001 and leave no pending writer. On main it answers Clear0000 from the stale bytes ("a sampled texture read its DCC keys before the recorded GPU write over them landed").
• pendingKeyStoreTests had no ctest entry, so it now runs alone as agc_pending_key_stores (--pending-key-stores-only), like the other recorder tests.

Related: #1359 (open) makes direct consumers wait for copied writes whose CPU write-back has not happened yet, DCC metadata included. This PR covers keys that recorded GPU work writes in place, which #1359 does not wait for. The two change different functions.

Overlaps with my other open PR; whichever merges second rebases:

• fix(agc): read comp-to-single DCC keys of sampled textures and storage images #4700 edits the line next to this one in textureClearKeys.

Tested

Windows 11 Pro 10.0.22631, AMD Ryzen 7 7800X3D, AMD Radeon RX 7900 XTX, WinLibs GCC 15.2.0, Release, BUILD_TESTING=ON, on main ffa1984. The branch was rebased onto main d539606 for this pull request and is not yet rebuilt or re-tested there:

• ctest: 952/952 passed: main's 951 (8 of them skip there too) plus the new agc_pending_key_stores.
• With DccMetadata.cpp as on main, agc_pending_key_stores fails with the message above.
• tools/check_conventions.py --base origin/main passes.
• Ready or Not (PPSA18605), from boot to the station: the [gpu] texture ... has mixed DCC keys ...; its texels are read as stored reports went from 11 and 14 (two runs without this change) to 2 in each of three runs with it. The 6 reported before the level load are gone.
• The two textures left are 2612x1472 RGBA8 (format 56) surfaces whose key bytes are mixed in guest memory. They are an open lead for the colour corruption reported in Ready or Not (PPSA18605) #4184.
• That was measured on an earlier state of the bring-up build: main bab7b32 plus fix(libkernel): restore Windows select build for Demons Souls testing #4021, fix(recompiler): include implicit inputs in relative vector windows #4057, feat(agc): render layered draws into 3D color targets #1077, feat(agc): depth planes sampled as 2D arrays, cubes and integer views #425, fix(libSceAgcDriver): retire a depth surface once another writer covers its memory #454, feat(shader): give reads the masked-select pass keeps the whole pixel ancillary word #4031, fix(libc): alias non-contiguous shared guest views on Windows #2254 and local changes, before perf(libc): remember clean pages once per run in Windows write collects #3864, perf(agc): clear integer-format storage images with DCC code 0000 on the GPU #3865, perf(agc): handle the zero dword fill kernel as a buffer fill #4012, perf(agc): sort guest buffer regions through indices before merging them #4248 and fix(agc): order direct reads after copied writes #1359 were added. It was not measured on this branch alone.

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 in TechnicalDebt
• 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: none
• AI-assisted: yes

[github-actions]

Other open pull requests touch the same code. Merging one will break the other, or both implement the same thing. If this one builds on them, list them in Depends on.

Pull request	Conflicting files (lines on main)	Same exports	Same new files
#4017	core/libs/prx/libSceAgcDriver/CMakeLists.txt L899
#4554	core/libs/prx/libSceAgcDriver/Graphics/src/DccMetadata.cpp L451-452

[Nesyth]

Closing in favour of #4554, see my comment there.