Support BFloat16 in full_like (pytorch#7822)

Partial fix for pytorch#7748.

Author: swolchok

kernels/portable/cpu/op_full_like.cpp

@@ -56,7 +56,7 @@ Tensor& full_like_out(
     CTYPE_VAL val;
     utils::extract_scalar(fill_value, &val);
 
-    ET_SWITCH_REALHB_TYPES(out_type, ctx, name, CTYPE_OUT, [&] {
+    ET_SWITCH_REALHBBF16_TYPES(out_type, ctx, name, CTYPE_OUT, [&] {
       CTYPE_OUT val_casted = static_cast<CTYPE_OUT>(val);
       auto data_out = out.mutable_data_ptr<CTYPE_OUT>();
       for (size_t i = 0; i < out.numel(); ++i) {

kernels/test/op_full_like_test.cpp

@@ -85,9 +85,9 @@ void OpFullLikeTest::test_full_like_out<ScalarType::Bool>() {
   EXPECT_TENSOR_EQ(out, tf.zeros(sizes));
 }
 
-TEST_F(OpFullLikeTest, AllRealOutputPasses) {
+TEST_F(OpFullLikeTest, AllDtypeOutputPasses) {
 #define TEST_ENTRY(ctype, dtype) test_full_like_out<ScalarType::dtype>();
-  ET_FORALL_REAL_TYPES_AND(Bool, TEST_ENTRY);
+  ET_FORALL_REALHBBF16_TYPES(TEST_ENTRY);
 #undef TEST_ENTRY
 }