[MarcZX10]

What

Fixes #6735. HoldPending treated any pending storage image as evidence that the just-submitted store for this surface had executed. Since submissions are asynchronous and earlier fixture cases can leave unrelated images pending, the helper could return before this surface had a live storage image; the subsequent ReadImage/FindLive check could then race the queue worker. It now waits for pending writes that overlap the surface under test.

Tested

• python3 tools/check_conventions.py --base upstream/main — passed.
• git diff --check — passed.
• agc_driver_store_over_untracked_tests target built on Linux after temporarily applying the independent missing-template compiler fix from fix(shader): qualify dependent member template #6738; that unrelated line was removed before commit.
• Focused CTest starts but aborts before the fixture runs: this host CPU lacks LZCNT, BMI1, BMI2, MOVBE, AVX2 and FMA.
• No GPU runtime result is claimed.

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; silent stubs are listed in technical debt
• No notes, investigation files or images added to the repository
• New third-party code is a submodule built from source
• Generic behaviour, not specific to one title
• Depends on: none
• Unblocks: none
• GPUs tested: n/a; no GPU runtime test was possible on this host
• AI-assisted: yes

[MarcZX10]

Closing in favor of #6744 after reviewing its updated evidence. The author reproduced the reported failure on unchanged main, showed that 37 of 44 accepted waits lacked a live or pending image for the tested surface, then passed 100/100 fresh Windows GPU runs on the exact committed fix. That is stronger and more specific than my earlier Linux-only build evidence, and the PR isolates the tested surface by address and size. My PR’s diagnosis overlapped; no separate fix is needed.