Consider new "raw" option for horizontal axis

This issue is motivated by an issue in current TensorFlow, which incorrectly annotates the step parameter for epoch_ summaries when using update_freq (https://github.com/tensorflow/tensorflow/issues/26873).

Anyway, I have a number of TF summaries that are impossible to compare because the x-axis values are meaningless. I can think of other examples where this may happen, for example, when using training sets of different sizes (let's say 100 and 1000 samples) and using different values of update_freq (e.g., =10 for the first and =100 for the latter) to get the same number of within-batch metrics saved.

The current x-axis options (step, relative [time], wall [time]) are not appropriate in these situations. Please consider an additional "raw" option that basically ignores the step values from the summary files.