Improve FPS propagation and metadata handling #42
Description
Currently, LISBET requires the user to provide the FPS (frames per second) value at multiple stages of the pipeline. This leads to unnecessary duplication and increases the chances of errors, especially when running different subcommands in separate sessions. The situation is even worse during inference, where we ask the user to manually provide the fps_scaling value, because we have no reliable way to retrieve the FPS used during training. This makes the overall behavior confusing and hard to explain.
The root of the issue is that we don’t have a proper way to store and propagate the FPS information once it is known. One way to fix this would be to rely more on xarray and adopt NetCDF (.nc) as the standard format for LISBET-generated data. We could still provide CSV exports for user-facing results, but internally we should aim to store data in a structured format that preserves both data and metadata.
In general, we should take this as an opportunity to review how FPS and similar metadata are handled across the pipeline, and implement a more consistent and centralized way to store and access this information. This would make the system easier to use and extend in the future.