vulkan: Enable topk_moe fusion for GLM-4.7-Flash

[jeffbolznv]

Just need to add the fusion detection logic, this is a combination of existing modes (early softmax, bias, norm, scale), and is covered by the existing backend tests.

before

Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo>llama-bench -m c:\models\GLM-4.7-Flash-Q4_K_M.gguf -r 10 -fa 1 -p 512 -n 128
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 5090 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo\ggml-vulkan.dll
load_backend: loaded CPU backend from Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo\ggml-cpu.dll
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| deepseek2 ?B Q4_K - Medium     |  16.88 GiB |    29.94 B | Vulkan     |  99 |  1 |           pp512 |      8434.22 ± 37.67 |
| deepseek2 ?B Q4_K - Medium     |  16.88 GiB |    29.94 B | Vulkan     |  99 |  1 |           tg128 |       185.26 ± 16.12 |

after

Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo>llama-bench -m c:\models\GLM-4.7-Flash-Q4_K_M.gguf -r 10 -fa 1 -p 512 -n 128
ggml_vulkan: Found 1 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 5090 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
load_backend: loaded Vulkan backend from Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo\ggml-vulkan.dll
load_backend: loaded CPU backend from Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo\ggml-cpu.dll
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| deepseek2 ?B Q4_K - Medium     |  16.88 GiB |    29.94 B | Vulkan     |  99 |  1 |           pp512 |      8504.07 ± 57.02 |
| deepseek2 ?B Q4_K - Medium     |  16.88 GiB |    29.94 B | Vulkan     |  99 |  1 |           tg128 |       206.38 ± 16.00 |

[jeffbolznv]

This may not be needed after #18980, I'll check once that lands.

[jeffbolznv]

I verified the model is hitting TOPK_MOE_SIGMOID_NORM_BIAS now, without this change. So we don't really need the change.

[0cc4m]

So this can be closed?

[jeffbolznv]

I'm fine with abandoning it. If another model needs it in the future the code will still be here.