add tvm-ffi metal

[ved1beta]

fixes #352

Summary

• Adds noarch variants for the tvm-ffi framework (tvm-ffi-metal, tvm-ffi-cuda, tvm-ffi-cpu, tvm-ffi-rocm, tvm-ffi-xpu, tvm-ffi-universal)
• Introduces TvmFfiNoarch mirroring TorchNoarch;
• Extends parse_variant and BUILD_VARIANT_REGEX
• Updates _sort_variants so within noarch, Torch is preferred over tvm-ffi

tests

Torch beats tvm-ffi when both are present
noarch fallback when no arch variant matches
arch variant beats noarch on the actual hardware

[danieldk]

We already have a branch for tvm-ffi and there is also support for kernel-builder in it. However Metal support will require upstream tvm-ffi changes, because there is currently no way for tvm-ffi to pass MTLBuffers (only pointers to the raw data), which is needed for other Metal APIs (as well as no way to get other compute context).

For this reason we are postponing tvm-ffi Metal support until these issue have been fixed upstream.

[ved1beta]

ahh thanks for the response , will try to put a pr upstream 🫡