V1 model inference with NAIP

[lillythomas]

The purpose of this PR is to create a notebook demonstrating how to run inference on new NAIP scenes. It may also serve the purpose of demonstrating sim search with the generated embeddings.

[lillythomas]

Some current examples of similarity search results generated with 3600~ embeddings from NAIP sized at 256x256 and the v0.5.3/mae_v0.5.3_epoch-08_val-loss-0.3150.ckpt checkpoint.

These are all from the Bay area of California in 2020. Kind of crazy that it even captures similarity between open ocean surfaces with specifically visible white capping.

The notebook in this PR can be run to reproduce.

[yellowcap]

@lillythomas I think this is a great example, and I think we should polish this up and add to the tutorials. I wanted to ask if you would be up for using stacchip for the data generation in the example 😊 . I made some small changes to stacchip to make it easier to use. Creating naip tiles is now just a few lines of code, see https://clay-foundation.github.io/stacchip/naip-tutorial/ .

[lillythomas]

@yellowcap sounds good! I'll work on swapping the method for getting NAIP chips to use stacchip and request your review on it.

[yellowcap]

Looks great overall! I did a few suggestions, mostly cosmetic changes here #287 .

A few more householding things

• Remove the non-stacchip example,
• Move the tutorial into the tutorials section of the docs (will need and update of the relative paths in the notebook)
• Run the notebook and commit the cells with output, but replace the image with a markdown cell that renders the image that you can upload to the PR. A bit hacky but keeps the repo small.

[yellowcap]

Almost there! Just a few small comments, mainly cosmetic things. Feel free to merge when updated.

• Register new notebook to _toc.yml
• add lancedb to pip install
• The cell where chips are downloaded is duplicated, remove older version
• Update relative paths to ../.. (notebook requires to levels down)
• Centroid extraction can be simplified:
  Replace

    lon, lat = box(
        bounding_box[0], bounding_box[1], bounding_box[2], bounding_box[3]
    ).centroid.coords[0]

with

    lon, lat = bounding_box.centroid.coords[0]