Add kerchunk opener to `repr` route

[cisaacstern]

The recently-added https://github.com/pangeo-forge/liveocean-feedstock had a successful production run:

https://pangeo-forge.org/dashboard/recipe-run/1391?feedstock_id=91

Our xarray repr generation route doesn't support kerchunk, so instead of a dataset repr the frontend is displaying:

status code: 404
An error occurred while fetching data from URL: https://api.pangeo-forge.org/repr/xarray/?url=https://ncsa.osn.xsede.org/Pangeo/pangeo-forge/pangeo-forge/liveocean-feedstock/liveocean.zarr
{"detail":"An error occurred while fetching the data from URL: https://ncsa.osn.xsede.org/Pangeo/pangeo-forge/pangeo-forge/liveocean-feedstock/liveocean.zarr. Dataset not found."}

Let's add some conditional logic for that rather than assuming all datasets are zarr, here:

https://github.com/pangeo-forge/pangeo-forge-orchestrator/blob/fb5caaf6994d1b73004b447ab00e5687ab941f2e/pangeo_forge_orchestrator/routers/repr.py#L24-L25

xref #198 #199

cc @rsignell-usgs this is why the frontend is not displaying the dataset repr for liveocean. But I think https://ncsa.osn.xsede.org/Pangeo/pangeo-forge/pangeo-forge/liveocean-feedstock/liveocean.zarr should be your full dataset (despite the inaccurate .zarr suffix, which is because of #199)?

[katamartin]

👋 gentle bump on this issue!

It looks like this is now affecting the first two recipes on the frontend, which makes it fairly prominent.

[cisaacstern]

Thanks for the ping @katamartin.

@andersy005, noticed your 👍, do you have interest in + bandwidth to work on this?

[andersy005]

@cisaacstern, this is on my TODO list, but won't be able to work on it until later next week :)

[cisaacstern]

Sounds great. I'll assign you here, @andersy005. From my perspective, no stress re: timeline. Of course lmk if you end up needing feedback/review on this.

[cisaacstern]

@katamartin, out of curiosity, does next/vercel offer the possibility of running any type of integration testing against a live instance of our backend application? For every PR, we have the option of deploying a review instance of the backend, and also merges to main deploy to a staging instance.

In #226, I am working on an integration test wiring these instances to Google Dataflow, as part of debugging #220. Similarly, should we consider some scheduled tests of the frontend against these dev/staging instances of the backend? I am imaging some future where we require checks such as:

✅ Dataflow integration passing
✅ Next.js frontend integration passing

Before we deploy to prod. (Anderson maybe you have some insight on how to do this too. When we spoke the other week, I think we agreed that frontend integration testing was not the highest priority, but here we have a prime example of when it would have helped already 😅 .)

[katamartin]

@cisaacstern ooh can definitely see how that would be useful here! Unfortunately, not sure that there are great out-of-the-box options.

As I see it, we would need to wire up deploys for the pangeo-forge.org project to this repo and get those frontend deploys to point to the backend preview deploy URLs. It looks like there's definitely no Vercel tooling for the latter (https://github.com/vercel/vercel/discussions/5231) and I'm not sure how well-supported the former would be either.