Show layout analysis for convolution ops

[sam-mccall]

This is designed to show how well your data fits the TPU. It looks like this:

The new data is described in a change to tensorflow/contrib/tpu/profiler/op_profile.proto that hasn't made it to github yet. This is what's added to XLA instruction info:

// Describes the physical memory layout of the instruction's primary input.
// e.g. for a convolution, this analyzes the image and ignores the kernel.
LayoutAnalysis layout = 5;
message LayoutAnalysis {
  // The physical data layout, from most-minor to most-major dimensions.
  repeated Dimension dimensions = 1;
  message Dimension {
    int32 size = 1;       // Size of the data in this dimension.
    int32 alignment = 2;  // Data must be padded to a multiple of alignment.
    string semantics = 3;  // What the dimension represents, e.g. "spatial".
  }
}