Support basic Nahual functionality

[afermg]

Nahual is a companion tool that allows us to deploy a multitude of models for segmentation (e.g., cellpose), tracking (e.g., trackastra) or featurization (e.g., Dinov2) in an independent Python environment or server altogether. I have integrated its functionality for multiple reasons:

• Relieve dependency contraints, such as baby, which requires older versions of many things. Now aliby can evolve independently from baby and still reuse it.
• Facilitate multiprocessing: Many issues arose when trying to use multiprocessing while using GPUs (e.g., cellpose breaks very quickly). Multithreading works to some degree, but then only a few CPUs are used. Decoupling the steps allows us to
• Scale up using multiple servers: If an aliby pipeline can contact multiple other processes in other computers it could make use of their GPUs/CPUs and process data much faster.

There is a bit of an overhead of moving data around, but at least for pipelines running within the same server it is not an issue (it may become if we start calling servers with low transfer speeds).