Dtype selective build: fail if not xplat, if portable/optimized not in kernel_deps (pytorch#11020)

Pull Request resolved: pytorch#10990

pytorch#10985

Try to make user error harder for dtype selective build.

Currently, if dtype selective build cannot be enabled we silently fallback to regular dtype selective build.

This diff:
1. Emits failure when there is nothing in kernel_deps and dtype_selective_build=True.
2. Remove dtype_selective_build=True for cases in (1).
3. Warning that dtype selective build is not enabled if user is not in xplat.

ghstack-source-id: 285201952

Differential Revision: [D75027794](https://our.internmc.facebook.com/intern/diff/D75027794/)

Author: pytorchbot

shim_et/xplat/executorch/codegen/codegen.bzl

@@ -641,6 +641,21 @@ def executorch_generated_lib(
     if not aten_mode and not functions_yaml_target and not custom_ops_yaml_target:
         fail("At least one of functions_yaml_target, custom_ops_yaml_target needs to be provided")
 
+    if dtype_selective_build:
+        if (not "//executorch/kernels/portable:operators" in kernel_deps) and (not "//executorch/kernels/optimized:optimized_operators" in kernel_deps):
+            fail("""
+            !!WARNING!! Dtype selective build is available for the portable and optimized kernel libraries.
+            If you are using those, please add them to `kernel_deps` in `executorch_generated_lib`:
+            //executorch/kernels/portable:operators
+            //executorch/kernels/optimized:optimized_operators
+            This will tell the build system to rebuild portable/optimized with the dtype selective build header.
+            For examples, see: //executorch/examples/selective_build/targets.bzl
+            Currently, kernel_deps contains {}.
+
+            If you have a custom kernel library, please remove `dtype_selective_build=True`
+            and use regular selective build.
+            """.format(kernel_deps))
+
     aten_suffix = "_aten" if aten_mode else ""
 
     # merge functions.yaml with fallback yaml