Prevent from building datasets which will cause a failure at runtime

[mihaibujanca]

Checks should be in place to validate .slam files before running experiments. This may either be done with a separate tool to check the validity of the .slam file, or could be part of each individual dataset.

List of possible things to check:
Same number of ground-truth measurements as the number of processed frames.
Number of RGB frames equal to number of depth frames.
For stereo: same number of frames in the left and the right camera.