Added the new type for the fixed size regressor.

Fixed some typos, the type was referring to the classifier instead
of the regressor.

Added the StatsBase dependency for sampling. Started working on
the training procedure for the fixed size regressor on a new
file to avoid cluttering.

First version of the training and prediction routines. Still working
out the logistics.

Some minor formatting on the docstrings.

Added a new keyword2dict utility for the fixed size regressor.

Updated the fields for the fixed size regressor to accept the subsample
size and the total number of iterations.

The constructor was missing a few arguments

Just formatting, also added a missing comma.

Added some comments in the training routine.

Always check if the solver has converged.

Added all the files and types to the main file. Still not working,
but a WIP.

We need those values unpacked.

Finished implementing both the training and the prediction functions.
Still WIP to see if it actually works.

Unnecessary code.

Added the regressor type for the MLJ interface.

Added the fitting step to the MLJ interface.

Added the prediction function for the fixed size regressor, using
the MLJ interface.

Working on the test sets for the new implementation.

All the prediction and training procedures work correctly.
The tests pass. Although, it seems to be quite time consuming.

Updated the tests, also replace the reshape statements for dropdims.
Updated the metadata as well.

Some tests were missing.

Some more information in the README.

Updated dependencies and fixed formatting.

A refactoring was needed to remove the problem with the naming. We could
still use the const, but I believe this is now best practice, even
though is might be quite verbose.

Some more name refactoring. Also, had to add the ScientificTypes
dependency to the project.

There was a bug with the scientific types for the classifier, which is
now fixed by implementing new trait methods.

All integration tests are now passing.

Subsampling and Nyström approximation is now done before fitting, with
the use of the `reformat` and `selectrows` methods.

For now, the FixedSizeRegressor is not working as expected when
reformatting the data. It is taking all the elements from the
transformations as arguments to the fit/predict methods.