[feat] Add new add-ons from the user side

[nabenabe0928]

Current Auto-pytorch already adapted to the addition of new components (e.g. lr_scheduler, optimizer) from user side, but it is not easy for them to know how it should be implemented for the following reasons:

1. not clear what variables, attributes or methods new add-ons should have
2. not clear how we can pass initial parameters to those add-ons

For those reasons, I opened this issue and would like everyone to put any ideas here.

[ravinkohli]

1. not clear what variables, attributes or methods new add-ons should have

We should definitely add examples for them in the documentation similar to auto-sklearn here

2. not clear how we can pass initial parameters to those add-ons

I'd say currently it can be done by using the hyperparameter updates, and maybe it's an easier way to pass the parameters even when we don't perform search.