docs(fold): open item 4 records TestMulGatherQMMGlobalScale's pass on Metal; BINARIES links the item

[glennneuber]

This fixes notes 1 and 2 from #404's review (comment). Docs only, two files.

• Open item 4. It said TestMulGatherQMMGlobalScale "is gated to Metal upstream and passes on CUDA with the gate widened". It now also says the test passes on Metal on the release build (docs(fold): Metal's vision goldens per model and the ./mlx tests, on the release build #404).
• The matching gate 4 bullet in the CUDA section gets the same clause: on Metal, where the gate lets the test run, it passes on the release build.
• Open item 5 cites docs(fold): Metal's vision goldens per model and the ./mlx tests, on the release build #404 beside docs(fold): v0.34.4 is deployed on the Apple Silicon host, with two-pass and f16 #402 for Metal's gate 4, since docs(fold): Metal's vision goldens per model and the ./mlx tests, on the release build #404 adds the per-model goldens and the ./mlx tests.
• BINARIES.md cited "(open item 4)", but that file has no open items. The citation now links the fold record's open items.

Notes 3 and 4 need the Metal host's data, so they are not in this PR: the 26b and 12b digests, and whether -count=1 ran.

Checks:

• check_source_paths.py --changed-since origin/main is clean.
• The name scan is clean.
• The BINARIES.md row keeps its seven column separators.
• The link's target and its #open-items anchor exist.

ai-server/mlx-cuda

🤖 Generated with Claude Code