Add Burger's dataset and PINO

[crwhite14]

This pull request adds the Burger's Equation dataset and adds PINO (with FDM) for FNO.
Note that it replaces the old PR that only added Burger's Equation.

Changes

• I changed the Burgers dataset to npz and created a new link using figshare.
• Added scripts/train_burgers.py
• Updated neuralop/datasets/burgers.py to load the new dataset, burgers.npy
• Created neuralop/losses, which contains data_losses.py, equation_losses.py, and meta_losses.py

How to run the Burgers dataset

• Download the dataset from here, and place it into /data/burgers/
• cd scripts
• To run FNO normally (with data loss): python train_burgers.py
• To run PINO (i.e., FNO with only equation loss): python train_burgers.py --config_name=default --config_file=burgers_pino_config.yaml

[dhpitt]

All looks good to me. Changes from the PINO branch's trainer are in #253

[JeanKossaifi]

Looks great @crwhite14 ! I'd just add a small PINO example so it's executed each time to show how to add equation loss (we can do in another PR) and rename the training script train_pino_burgers so it's clear it has equation loss.

[JeanKossaifi]

Merging.