Add `all` transforms to train script

[dakinggg]

In https://github.com/mosaicml/llm-foundry/pull/1290, I added a unit test for the all transform, but forgot to add it to train.py

Loss curve before and after, also verified the batch size calculations are done the same.