[SM6.10] Fix numeric types allowed in LinAlg intrins

[V-FEXrt]

Fixes #8271

Use LinAlg as the intrinsic type instead of numeric for LinAlg intrinsics and update the LinAlg to match current allowed types

[tex3d]

I actually don't think we want/need to keep AR_BASIC_LITERAL_INT, or AR_BASIC_LITERAL_FLOAT here.

Also, AR_BASIC_FLOAT32_PARTIAL_PRECISION is half when in min-precision mode that maps to ordinary float. I don't think it's necessary to support that, but it probably doesn't hurt.

[V-FEXrt]

fix will be up in a couple minutes

[V-FEXrt]

Okay so removing AR_BASIC_LITERAL_INT caused some test failures. Specifically, in SetElement

dxc/DirectXShaderCompiler/tools/clang/test/CodeGenDXIL/hlsl/linalg/builtins/matrixsetelement/nominal.hlsl:21:3: note: candidate function not viable: no known conversion from 'literal int' to 'unsigned int' for 4th argument
  __builtin_LinAlg_MatrixSetElement(mat2, mat1, 1, 5);

I think this would be fine in real use cases because of the header implementation but given that is a slightly larger change/longer discussion I'm going to take your offer to punt :)

[tex3d]

We could address tightening of the g_LinAlgCT in the future. For now, this looks good.

[V-FEXrt]

#8468 issue to track follow up fixes/discussions