vulkan: Allow non-pow2 n_experts in topk_moe

[jeffbolznv]

Saw granite-3.0-3b-a800m-instruct-Q8_0.gguf being used at https://www.phoronix.com/review/llama-cpp-vulkan-eoy2025/3, with lower than expected scaling for 5090.

before:

Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo>llama-bench -fa 1 -p 512 -n 128 --prio 1 -r 10 -m c:\models\granite-3.0-3b-a800m-instruct-Q8_0.gguf
ggml_vulkan: 0 = NVIDIA GeForce RTX 4070 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           pp512 |    11545.75 ± 199.63 |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           tg128 |        305.26 ± 3.35 |

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
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           pp512 |   26586.25 ± 3416.45 |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           tg128 |        455.17 ± 6.23 |

after:

Z:\github\jeffbolznv\llama.cpp\build\bin\RelWithDebInfo>llama-bench -fa 1 -p 512 -n 128 --prio 1 -r 10 -m c:\models\granite-3.0-3b-a800m-instruct-Q8_0.gguf
ggml_vulkan: 0 = NVIDIA GeForce RTX 4070 (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           pp512 |    11704.54 ± 179.34 |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           tg128 |        325.64 ± 1.62 |

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
| model                          |       size |     params | backend    | ngl | fa |            test |                  t/s |
| ------------------------------ | ---------: | ---------: | ---------- | --: | -: | --------------: | -------------------: |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           pp512 |    28154.26 ± 504.16 |
| granitemoe 3B Q8_0             |   3.34 GiB |     3.37 B | Vulkan     |  99 |  1 |           tg128 |       521.35 ± 10.23 |