There was common situations where two provider instances where running
in parallel. And was creating weird errors (like getting wrong api_uri)
because the bootstrap deferreds were global but the Provider objects
not.

I don't like much singletons, but I think now is simpler than before.

- Resolves: #9073

whoever wrote this in the first place needs to get introduced to
import-time initialization, even if it was me. one of the nastiest
python gotchas in my opinion :)

just to make the logic clear.
we should probably check that the file is not empty and it has valid
json (sanitized against the spec).

here we port the if-modified-since conditional mechanism, so that we
only write the config if it is newer than whan we have.

we also add a line with the status code to the logs, so that it's easier
to debug.

note that the 'configs.json' file is never returning 304.

- Resolves: #8773

The workaround is no needed anymore, if we do 'GET' instead of 'POST' to
fetch all the provider configuration jsons it works smoothly.

- Resolves: #9056

Old versions of the webapp let agent download an invalid.json, instead
of returning a 404. We try to parse the json, and if no valid json is
found, we use the workaround for manually downloading the service files.

- Resolves: #9004

if there is no ca_cert passed and the provider is configured, use the
pinned ca cert when the autoconf parameter is true.

leap.common is going to handle this

- Resolves: #8881