add more parameters to control fitting, and add data checks

[OkonSamuel]

This PR add several checks on the dataset. One of such checks is to make sure that the dataset doesn't contain more features than observations (i.e the dataset isn't wide). This PR also throws an informative error message for some hidden errors due to flawed offsetcol logic. For example, users were previously allowed (maybe unintentionally) to choose an offset column that wasn't among the features of the training dataset, leading to confusing errors.

cc. @ablaom , @rikhuijzer

[OkonSamuel]

I feel that using @formula syntax here is unnecessary as currently one has to

1. materialize a matrix, append the target
2. convert the appended matrix back to a table
   then pass to the GLM lm/glm call which
3. Materializes the appended matrix.
4. effectively splits the target vector from the feature_matrix.

I find it unnecessary to use @formula syntax when we could just have done (1) above then passed it to the GLM lm/glm call which doesn't need to perform (3) and (4) above.
Although the main advantage of the @formula syntax is it's flexibility and the ease of use it provides the user in selecting combination of features to fit on, this doesn't apply here as the @formula syntax isn't currently exposed to the end user. Also for simple selection of features MLJ has several Transformers to do this.
As mentioned in JuliaAI/MLJModels.jl#406 (comment), the @formula syntax could make transformers easier to use.
So the @formula syntax should be dropped from this package as soon as JuliaAI/MLJModels.jl#406 (comment) is resolved.

[rikhuijzer]

I'm sorry for the late reply. I forgot about this PR.

I feel that using @formula syntax here is unnecessary
[...]
this doesn't apply here as the @formula syntax isn't currently exposed to the end user.

Hmm yes good point I guess.

[ablaom]

Thank you @OkonSamuel for this extend sequence of PRs and @rikhuijzer for the informed reviews. ❤️