vulkan: Update topk_moe fusion to handle gpt's late softmax

[jeffbolznv]

Based on #16649.

[jeffbolznv]

CC @am17an I've included the ggml_check_edges change in this PR.

[0cc4m]

I understand what this change is doing, but how do I test it? The topk_moe tests pass before and after this change. Which model architectures correspond to the three modes?

[0cc4m]

Vulka?

[jeffbolznv]

This logging is from ggml_backend_sched_print_assignments (set GGML_SCHED_DEBUG=2, run llama-bench with -v, I also edit the code not to skip views), it truncates the name to keep the formatting aligned.

[am17an]

I understand what this change is doing, but how do I test it? The topk_moe tests pass before and after this change. Which model architectures correspond to the three modes?

Usually I put a debug statement printing the number of nodes fused. We'll need to come up with a better way to assert that the nodes were actually fused

[jeffbolznv]

I understand what this change is doing, but how do I test it? The topk_moe tests pass before and after this change. Which model architectures correspond to the three modes?

I've added some logging in the latest commit that I use to verify fusion and the effects of graph_optimize. You can see the whole sequence of ops without a sync in between, which implies the fusion is working.

Early softmax w/norm: qwen3
Early softmax w/o norm: deepseek2
Late softmax: gpt-oss