Generalize data paths for any data source

[lalalune]

This is a PR to generalize handling of data sources. Currently the code makes assumptions that users will be using one of the provided training sets, but most likely people will want to evaluate on other training data or other parts of ShapeNet.

Changes the following things:

1. Removes if statements and dataset-specific logic
2. Change data_camera_mode to manifest_dir -- the main function is to find the training and val text files once any specific logic has been generalized.
3. Assume use_shapenet_split to always be true and remove the arg -- the other PR adds a script which properly splits data into 70/20/10 file manifests.
4. Update README to reflect changes (and fix a few typos also)

Submitted as a draft, I am testing on some datasets now to make sure everything is working.

[lalalune]

Was able to get training running successfully with this, un-drafting

[dmoham1476]

When will the google colab be updated with this PR?

[lenashamseldin]

Hello, could you please update on how you got training to work on your own data? What did you change in the parameters of the commands and also did you need to add the path to obj files as well? Or only images? I copied your edited files and replaced them in the files open on colab but it doesn't seem to see any images in the folder path I passed...Would appreciate your help:)

[dav-ell]

Also looking for an update on this. These features are pretty critical to using this repo.