add empirical model to ngiab

[JoshCu]

uses the new ngiab branch of ngen than contains the model in the /extern/ folder CIROH-UA/ngen#3 CIROH-UA/ngen@8e8fdcd

The extra index url line makes sure pytorch is installed as cpu only because the gpu packages / drivers make the image 6gb larger.

The package needs to be installed with UV after troute and ngen so that the numpy version matches.

I've tested this with some example data, the preprocessor will generate the correct configs if you add --em to it.

python -m ngiab_data_cli -i gage-10154200 --start 2010-10-01 --end 2010-11-01 -o lstm_test -sfr --em

you can also run the bmi unit tests inside the image by entering the bash shell in the container and

cd /ngen/ngen/extern/lstm/
python -m lstm

[benlee0423]

I tested the sample data with merge image, and it worked both serial and parallel run.

[benlee0423]

@JoshCu
Is lstm integration done with this PR?
Do you have more commits or another PR after this?

[JoshCu]

@JoshCu
Is lstm integration done with this PR?
Do you have more commits or another PR after this?

This pr fully integrates the lstm, it was wonderfully simple thanks to how the repo is setup. I've referred to it as the empirical model after talking to the original authors because it's an older ml model that doesn't have all the latest improvements and performance of newer models. Hopefully it will make it easier to distinguish between this model and the latest lstm models being developed. It's referred to as the empirical model in the preprocessor also.