feat(moore): call MATE flash attention natively

[voltjia]

Summary

• Replace the per-call Python/MATE bridge from feat(moore): support flash-attn via MooreThreads/mate v0.2.5 #983 with a TVM-FFI native execution path.
• Use MATE only once per operator to resolve/compile the selected FMHA and combine modules, then cache their loaded DSOs and __tvm_ffi_* entry points.
• Write FlashAttention output and LSE directly into InfiniOps-owned tensors, removing provider allocations and D2D result copies.
• Route the Moore MATE glue through the existing system-C++ Torch object path so the vendor MUSA compiler does not parse ATen headers.

Validation

• Moore minimal linked build.
• Existing FlashAttention varlen/KV-cache correctness suites.
• Repeat-call check confirming Python is bootstrap-only.
• Non-default stream and split-KV combine coverage.

This is a draft while Moore hardware validation is in progress.

[voltjia]

Fix-up for the Moore build/runtime failures in \19c6490d:\n\n- Load TVM-FFI's DLPack header before ATen's vendored DLPack header, so the TVM-FFI ABI sees \DLManagedTensorVersioned\ and the current \DLTensor\ layout.\n- Keep ATen/MATE runtime state in each .cc\ via a forward-declared \MateState; generated MUSA dispatch units therefore only consume a lightweight provider declaration.\n- Use PyTorch's official \�t::Tensor\ pybind caster, mark the MATE loader lambdas mutable, and convert external softmax-LSE views before handing them to MATE.\n\nValidated on the new Moore S5000 host using Torch 2.7.1, MATE 0.2.5, and \�pache_tvm_ffi 0.1.9.post3+musa.1. I first built current InfiniRT master with \WITH_CPU=ON -DWITH_MOORE=ON, then built this PR with \WITH_LINKED=ON\ and the two FlashAttention operators selected. Compilation and linking passed.\n\nExisting tests, no new smoke tests: \pytest tests/test_flash_attn_varlen_func.py tests/test_flash_attn_with_kvcache.py --devices moore -k 'musa-16 and (q_lens0 or dense or paged)'\ -> 13 passed, 5 skipped, 94 deselected. The skipped cases are existing capability-specific skips.