Enforce raw/derived restrictions in Dataset

[jl-wynen]

Currently, Dataset holds the union of fields of RawDataset and DerivedDataset and allows users to set all, regardless of the selected type. Should we instead allow setting only the fields of the selected type? (Note that type is mandatory in __init__, so it is always know. In particular if #146 gets implemented.)

This would help avoid the confusion between investigator and principalInvestigator.

This would have to be enforced in:

• setters
• __init__
• derive
• replace

Problem: it is possible to have invalid datasets in SciCat. We should still be able to represent those in Scitacean (#134). This would require a backdoor in the validation in order to create such a Dataset instance. And there already are backdoors. I don't want to add more.

[YooSunYoung]

Will it be too complicated to have type and flexibility...?
For example, we may have two types: raw and derived and two flexibility strict and flexible...?

Another thing I thought was to have read-only-Dataset, which can be used loading/downloading dataset and it should be flexible since they may read invalid datasets.
And writable-Dataset can strictly forbid invalid fields.
Because invalid Dataset will be forbidden to upload anyways right...?