Improve the priority system

[krassowski]

When adding pyright integration a simple priority system was implemented to allow testing multiple servers on CI. There are still things to do:

• [x] Release blocker: reduce number of tie messages, do not consider servers for use if not available (something is wrong on binder?)
• [ ] Make the settings reload connections if needed to reflect the priority changes - maybe just show message prompting for page reload for the closest release?
• [ ] Implement per-document type/path priority settings as originally proposed by @bollwyvl in https://github.com/jupyter-lsp/jupyterlab-lsp/issues/587#issuecomment-829343095:

yeah, a relatively quick band-aid that wouldn't require much spec rework would be to land it in user settings. Is this a place to start the idea of a regex-based approach for e.g. using different servers for notebooks and python files?

{
    "language_servers": {
      "python-lsp-server": {
         "priority": 100,
      },
      "pyright": {
         "priority": 1000,
         "serverSettings": {
            "python.analysis.useLibraryCodeForTypes": true
         },
         "_path_overrides": {
            "priority": {
               "ipynb$": 0
            }
         }
      }
   }
}

In this case, serverSettings would not be overrideable by path (since we would still just have one language server process per implementation).

This would give us a fair amount of flexibility until we figure out how to do multiple simultaneous sources properly, which is really the path forward...