[ENH] Rework architecture specific to BIDS and CAPS dataset

[NicolasGensollen]

Closes #1448

Following #1443, this PR continues the re-organization of the code relative to iotools and converters. More precisely, it targets the code handling BIDS and CAPS datasets which is scattered across the code base (some in clinica.converters, some in clinica.iotools, some in clinica.utils...).

The base idea of this PR is to regroup all this code in a separate module called clinica.datasets (name can change if needed). This module has all the logic related to BIDS and CAPS datasets (validation, queries, writing, parsing...).
The logic specific to BIDS datasets is in the clinica.datasets.bids submodule while the logic specific to CAPS is in clinica.dataset.caps. Logic handling both is in clinica.datasets.

Going through these pieces of code, it seems we have a lot of small functions doing more or less the same things, and sometimes in ways that we don't need anymore. An example is the functions trying to guess whether a folder is a BIDS or CAPS dataset by doing some heuristic checks. Since the introduction of the MANDATORY dataset_description.json file for both types of datasets, we don't need to guess anymore, we have the information written in the metadata, so we "just" need to parse it.

[NicolasGensollen]

Alright, looks like all tests are passing again.
@AliceJoubert sorry again for the large PR.
Let me know what you think when you have some time to make a review. Again, I've made some choices but we can discuss them if you feel there are better alternatives.