Repin DiffusionGemma to the commit that merges cleanly

[danielhanchen]

Repin ggml-org/llama.cpp#24423 (DiffusionGemma) from c6f8d604 to 168b21ad.

Why

The publishing run 34108333979 failed in Resolve tag:

ggml-org/llama.cpp#24423 (c6f8d604b67611b73f7965c0bd39d26e7365a489) does not merge
cleanly onto b10830 + the PRs listed before it

Same shape as #205, different pin. The base moved from b10825 to b10830 and a
second pinned PR collided with the newer upstream tree, in the same file both times:
tests/test-llama-archs.cpp.

The conflict is one hunk, and additive_merge.py was right to refuse it, because the
base line already existed and both sides edited it:

base:      COHERE2MOE || MIMO2 || STEP35 || MUSE_GLIMMER || GRANITE_SWA || DOTS3NOTE
ours:      ... || DIFFUSION_GEMMA          (plus the comment explaining why)
upstream:  ... || SPARK2_5

Union of the two additions, keeping the DIFFUSION_GEMMA comment. 168b21ad is that
merge on the PR branch; ggml-org#24423 has gone from CONFLICTING to
MERGEABLE.

Every pin checked against the new base this time

Rather than fix one and rediscover the next on the following run, I merged all
thirteen pins onto b10830 individually:

pin	onto b10830
c6f8d604 ggml-org#24423	refused (fixed here)
36df1bf4 ggml-org#25731	resolves additively
b9b8207f ggml-org#27754	clean (the #205 fix still holds on the newer base)
the other ten	clean

So this should be the last blocker for the set.

Verification

Against the resolved tree: libllama.so and llama-common build clean, and
test-llama-archs builds and exits 0.

Note on the pattern

This is now twice in two days, both in tests/test-llama-archs.cpp, both "upstream
added an architecture to a list while one of our pinned PRs added its own". It will
recur every time the base advances past a pinned PR that touches that list. Worth
considering separately whether additive_merge.py should handle a disjunction or
comma list where the merge base is non-empty but each side only appended a distinct
term; I have deliberately not changed it here, because loosening a guard that
currently refuses to guess is a bigger decision than unblocking a release.

[chatgpt-codex-connector]

You have reached your Codex usage limits for security reviews. Please try again later.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-09-07T10:03:45.820629Z	90b9dc8	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.