[feature] implement the comet ml backend (and easy to switch to it)

[nyk510]

vivid implements experiment backend as a class for storing the log during learning. The only real implementation of this is currently a local backend that stores data to local.

There are many ways to log other than local storage, so we'll implement them. The first is comet_ml

NOTE

• Unlike local storage, log services that post data to the server, such as comet ml, need to explicitly indicate the start and end of the learning. Since we don't currently have that feature, we'll implement it by wrapping the call with a with clause or something like that.
• Currently, there is no support for backend switching (it creates an instance of local experiment backend directly in the class). Remove this and allow to take an experiment as an argument when fitting (by default, using local experiment).