Allow softmax and log_softmax to operate on any dimension (#5694)

Summary:
Pull Request resolved: #5694

Use the portable implementation for `dim != -1`, simplified to only floats.

Make the optimized cases fallback to the CPU impl.

Reviewed By: manuelcandales

Differential Revision: D63430960

fbshipit-source-id: ec2d56a60d2b064744383a2f1266dee7600fa884

Author: mcremon-meta

kernels/portable/cpu/util/targets.bzl

@@ -205,7 +205,7 @@ def define_common_targets():
             "//executorch/runtime/core/exec_aten/util:tensor_util",
             ":broadcast_util",
         ],
-        visibility = ["//executorch/kernels/portable/cpu/...", "//executorch/kernels/quantized/..."],
+        visibility = ["//executorch/kernels/portable/cpu/...", "//executorch/kernels/quantized/...", "@EXECUTORCH_CLIENTS"],
     )
 
     runtime.cxx_library(
@@ -253,5 +253,6 @@ def define_common_targets():
                 "//executorch/extension/llm/custom_ops/...",
                 "//executorch/kernels/portable/cpu/...",
                 "//executorch/kernels/quantized/...",
+                "@EXECUTORCH_CLIENTS",
             ],
         )