clarify 3 things in docs

[williamFalcon]

clarify 3 things in docs

1. DP, DDP don't matter with TPUs. TPUs work in DDP by default (at least rn).

2. Explain how the different steps are called (i thought we did).
   (here: https://pytorch-lightning.readthedocs.io/en/0.7.1/lightning-module.html#training-loop-structure)

# how val loop happens


all_outs = []

for batch in val_data:
    batches = split_batch(batch)
    outs = []
    for batch in batches:
         
         # call 1
         out = validation_step()
         outs.append(out)
    
    # call 2
    batch_out = validation_end(outs)
    all_outs.append(batch_out)

# call 3
validation_epoch_end(all_outs)

3. Explain that checkpoint + early stopping don't get overridden by things passed in as callbacks.
   (here: https://pytorch-lightning.readthedocs.io/en/0.7.1/callbacks.html)