musa: enable fp16 mma (all) and cublas on qy2

[yeahdongcn]

Make sure to read the contributing guidelines before submitting a PR

This PR is a rework of #13149.
Regarding the cublasGemmBatchedEx issue on the MTT S4000, I will continue investigating it with our MUBLAS team to identify a proper solution.

Testing Done

All tests below were performed on the MTT S80.

• Build completed successfully
• ./build/bin/test-backend-ops passed

  root@eab90abe42cc:/ws# ./build/bin/test-backend-ops 
  ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
  ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
  ggml_cuda_init: found 1 MUSA devices:
    Device 0: MTT S80, compute capability 2.1, VMM: yes
  Testing 2 devices
  
  Backend 1/2: MUSA0
    Device description: MTT S80
    Device memory: 16297 MB (15723 MB free)
  
    ABS(type=f16,ne_a=[128,2,2,2],v=0): OK
    ABS(type=f16,ne_a=[5,7,11,13],v=0): OK
    ...
    CROSS_ENTROPY_LOSS(type=f32,ne=[10,5,4,3]): OK
    CROSS_ENTROPY_LOSS(type=f32,ne=[30000,1,1,1]): OK
    CROSS_ENTROPY_LOSS_BACK(type=f32,ne=[10,5,4,3]): OK
    CROSS_ENTROPY_LOSS_BACK(type=f32,ne=[30000,1,1,1]): OK
    OPT_STEP_ADAMW(type=f32,ne=[10,5,4,3]): OK
    5527/5527 tests passed
    Backend MUSA0: OK
  
  Backend 2/2: CPU
    Skipping CPU backend
  2/2 backends passed
  OK

• Tested DeepSeek-R1-Distill-Qwen-7B-Q4_K_M.gguf, qwen3_8b_q4_k_m.gguf, nvidia-llama-3_1-nemotron-nano-8b-v1-q4_k_m.gguf with or without the -fa flag

[yeahdongcn]

Let's hold off on merging this until I upgrade to the upcoming MUSA SDK and verify that everything works as expected.

[yeahdongcn]

Rebased onto upstream/master.