Fix selective build codegen for sym primops

Summary:
Build-time codegen for selective build fails for symint/symfloat/etc. prim ops, for example `"executorch_prim::add.Scalar": ["v1"]`. At appears that the codegen logic doesn't currently handle kernels with no dtypes.

This change was split from D63882723.

Differential Revision: D64883035

Author: GregoryComer

codegen/tools/gen_selected_op_variants.py

@@ -72,7 +72,7 @@ def write_selected_op_variants(yaml_file_path: str, output_dir: str) -> None:
         for operator_name, kernel_metadata_str in et_kernel_metadata.items():
             tensor_meta = []
             for kernel_metadata in kernel_metadata_str:
-                if kernel_metadata == "default":
+                if kernel_metadata == "default" or "/" not in kernel_metadata:
                     break
                 else:
                     x = kernel_metadata.split("/")[1]