PR tensorflow#6697: fix build failure of run_hlo_module and interacti…

…ve_graphviz

Imported from GitHub PR openxla/xla#6697

cuda_executor references command_buffer here:
https://github.com/openxla/xla/blob/main/xla/stream_executor/cuda/cuda_executor.cc#L426

tsl::Status GpuExecutor::Submit(Stream* stream,
                                const CommandBuffer& command_buffer) {
  if (command_buffer.mode() != CommandBuffer::Mode::kPrimary) {
    return absl::InvalidArgumentError(
        "Can't submit non-primary command buffer for execution");
  }

  auto exec = GpuCommandBuffer::Cast(&command_buffer)->executable();
  VLOG(3) << "Launch command buffer execuable graph " << exec
          << " on a stream: " << stream->DebugStreamPointers();
  return GpuDriver::GraphLaunch(exec, AsGpuStreamValue(stream));
}

There are some missing dependencies declarations
Copybara import of the project:

--
4eb67bf5b16319f93594a7784a546bf8d04a5c9f by Shawn Wang <shawnw@nvidia.com>:

fix build failure of run_hlo_module and interactive_graphviz

Merging this change closes tensorflow#6697

PiperOrigin-RevId: 579106507

third_party/xla/xla/stream_executor/cuda/BUILD

@@ -524,6 +524,7 @@ cc_library(
         "@com_google_absl//absl/status",
         "@com_google_absl//absl/strings",
         "@com_google_absl//absl/strings:str_format",
+        "//xla/stream_executor:command_buffer",
         "//xla/stream_executor:plugin_registry",
         "//xla/stream_executor:stream_executor_headers",
         "//xla/stream_executor:stream_executor_internal",