[WIP] gui: move network calls to a dedicated process

[benoit-pierre]

To avoid freezing the GUI.

Notes:

• the worker process is forked, but a new interpreter is then started to ensure a clean slate: I could not find a way to dynamically require Gtk et al. only in the main process.
• the worker process higher priority requests first (FIFO order).
• the worker still process only one request at a time, so if for example the Invidious instance is down, a timeout of the current request can take a while and delay other (including higher priority) requests.
• the thumbnails are now independently saved to disk in ~/.cache/pipe-viewer.
• the results list is now cleared before sending a new search request (instead of before displaying the results), rational: there's no much point in being able to keep interacting with the current results as most search request should be fast.

TODO:

• [ ] get rid of the couple of remaining network requests in the main process (when saving a channel).
• [ ] get rid of the WWW::PipeViewer instance ($yv_obj) in the main process? Moving some of the helper methods still used to WWW::PipeViewer::Utils would make the separation between the part that needs online access and the offline stuff clearer.
• [ ] get rid of the annoying details window layout shift when filling the extra info.

Thoughts?