Move stateful initialization to a pre_fit function

[timokau]

Description

Creates a _pre_fit function to bridge the gap between __init__ and fit: Everything which initializes some state, but does not actually depend on the data, belongs there. This separates concerns a bit and makes it possible to inherit those initializations.

I thought about making this an implicit wrapper (a wrapper defined in learner that inherits signature & docs and automatically prepends a call to pre_fit to every fit call). This might be possible with wrapt and a meta-class, but it would be a little complex. Since it would also be less explicit, it might be a bit confusing for people who are not very familiar with the code. So I just went with an explicit call to _pre_fit for now.

I could imagine to automate the calls to initialize_optimizer and initiialize_regularizer in the future, but again I'm not entirely certain its desirable because it would feel a little "magic".

Motivation and Context

#157 (comment)

How Has This Been Tested?

Pre-commit & test suite.

Does this close/impact existing issues?

Impacts #94, #116, #146.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to change)

Checklist:

• My code follows the code style of this project.
• My change requires a change to the documentation.
• I have updated the documentation accordingly.
• I have added tests to cover my changes.

[kiudee]

Looks like a very clean solution. I like it.