[metal] Add MetalBackend method overrides for codegen

[aditvenk]

Stacked PRs:

• [metal] Auto-cap threadgroup size via CuteND/CuteFlattenedTileStrategy #1855
• Add backend-agnostic lane loop APIs to tile strategies #1798
• [metal] Add elementwise kernel tests #1796
• Add Metal test job to CI test matrix #1862
• [metal] Add MSL AST walker for Python-to-C++ translation #1794
• Add Backend.post_process_function_def hook #1851
• [metal] Add Metal codegen handlers for load, store, and mask_to #1854
• [metal] Reuse Inductor's MetalOverrides for MSL expression emission #1853
• ->[metal] Add MetalBackend method overrides for codegen #1852

---

[metal] Add MetalBackend method overrides for codegen

Add MetalBackend overrides for index_type, constexpr, program_id,
grid_index, full/reshape/broadcast_to/zeros expressions, where/min,
autotune, transform_host_arg, and launcher_keyword_args.

[jansel]

Is there test coverage for this? Maybe it would be good to get things working to the point you can compile a basic add kernel.

[jansel]

I don't think this is right.

[aditvenk]

Metal uses per-thread scalar dispatch — each thread processes one element, so there are no block dimensions. Hence, these are identity operations. Tests coming up in #1794 and #1796.

I also have matmul tests in an unpublished PRs on top of this current stack. Matmuls will use a cooperative thread model, similar to cute

[aditvenk]

Is there test coverage for this? Maybe it would be good to get things working to the point you can compile a basic add kernel.

Tests for copy kernels are coming in later PR in the stack: #1794
Tests for elementwise are also in the same stack: #1796

I broke up 1794 in smaller PRs for easier reviewing.