Investigate using ftw-inference-api as option instead of local only

[cholmes]

Running inference can often take a lot of time on a local machine. Once the ftw-inference-api is fleshed out and stood up it could be interesting to use the QGIS interface but have it drive the API instead of calculating things locally. If it's on a nice GPU machine it could go decently faster.

[m-mohr]

I'd recommend implementing it in a way that it allows both options.
If the API ever gets decomissioned, the whole plugin would go unusable and then having the local alternative would be great.
So either determine automatically whether the API is still up, or let a user choose whether they want to run on the API or locally.
Let me know if you need any help @Gedeon-m-gedus, but for now I'm unassigning me because I'm not sure what my todo is in this task.