Prebuilt: repin Inkling after merging b10229 into the branch

[danielhanchen]

Last night's schedule (30767267935) failed in Resolve tag:

ggml-org/llama.cpp#25731 (02142bbc33d477880a89dc5854a2c611d36e3494) does not merge
cleanly onto b10229 + the PRs listed before it; reorder or drop it in
scripts/unsloth/pr-set.json

This is not the pin rot that broke 07-31 and 08-01. All four pins passed the membership gate this time, and each was still at its PR head. Straightforward upstream drift instead.

Cause

Between b10226 and b10229, upstream added LLM_ARCH_DEEPSEEK4 to the same fallthrough group in src/llama-arch.cpp that Inkling registers in:

        case LLM_ARCH_QWEN35:
        case LLM_ARCH_QWEN35MOE:
<<<<<<< HEAD
        case LLM_ARCH_INKLING:
=======
        case LLM_ARCH_DEEPSEEK4:
>>>>>>> b10229
            return true;

The resolver merges pins onto the base tag and has no conflict resolution, so it refused. b10225-mix-345e1e3 published fine yesterday because the base was b10225, before DEEPSEEK4 landed.

Fix

Resolved on the branch, not here. b10229 is merged into add-inkling and both cases are kept, since both arches belong in that group. That is the last-writer problem this repo keeps hitting: whoever merges second has to know about the first.

The pin moves 02142bbc -> 1e6f9e4a. Nothing else changes:

PR	pin	change
ggml-org#24423 DiffusionGemma	c3fb9724	unchanged
ggml-org#25731 Inkling	1e6f9e4a	was 02142bbc
#48 Kimi-K3 fixes + MoonViT-3d	daef2b3e	unchanged
ggml-org#26185 Kimi-K3 text	04d6828b	unchanged

#48 needed nothing. It already carries b10223, and DEEPSEEK4 does not land anywhere Kimi-K3 registers.

Side benefit: ggml-org#25731's own mergeable state against master had gone red for exactly this conflict. It is back to true.

Verification

Replayed the resolver's merge sequence against b10229:

OK   c3fb972412  (24423)
OK   1e6f9e4a59  (25731, new pin)
OK   daef2b3e1b  (48)
OK   04d6828b2b  (26185)   <- still a no-op
UNMERGED PATHS: 0

Merged tree builds clean with -DGGML_CUDA=ON -DLLAMA_BUILD_TESTS=ON. test-chat passes; test-llama-archs -a kimi-k3 is OK on CUDA and CPU (NMSE 8.77e-08).

The new Inkling head also merge-trees clean against upstream master, which is six commits past b10229, so tonight's tag should resolve without another repin.