Projector does not validate tensor shape

There are a few fantastic examples of how to write data for consumption in the projector (e.g. https://stackoverflow.com/a/41370610 ). I tried applying these examples to my own work only to find that my checkpoints would not load. It turns out that the error was due to the shape of tensors in the checkpoint file; the tensor selected for the Project must be of shape N x d (for N examples of vectors with size d) or else the Projector silently fails. This issue is a request to the Tensorboard authors to add some sort of error message, tooltip, help message, etc. that might indicate to the user that a checkpoint was loaded but no viable tensors were found.

The issue names no files or tests, so begin by tracing the Projector's checkpoint-loading and tensor-selection flow, focusing on tensors that are not N × d. Confirm where the current silent failure occurs. Done means users receive a clear indication when a checkpoint loads but contains no viable tensors, with coverage for the invalid-shape case.