Load-confounds

This is the first beta release of the `load_confound` project. Updates include:
* code refactoring. The code (and the tests) are more modular, and each module is kept short. This refactoring impacted the `compcor` method in particular.
* support for gifti surface data files.
* checks and sensible error messages to match input files with specific strategies. Note that it is not possible anymore to directly specify a `tsv` confound file, as the target imaging file is necessary to assess that a particular strategy is sensible.
* better documentation for ICA AROMA.
* full redesign of the denoising strategies. Instead of following a particular benchmark, the denoising strategies are now based on the major technical variants used in the literature, and are quite customizable.

Congratulations to all contributors for reaching the first beta release, and special thanks to htwangtw for all the hard work that went specifically into that new release :tada: :tada: :tada:

The next release will improve support for 'hard' scrubbing, and will likely introduce a new API. The `load_confounds` project is also set to merge inside [nilearn](nilearn.github.io/) and will be discontinued as a stand-alone repository when the merge is complete.

This major update (0.11.1) is brought to you by [Brainhack Global 2020](https://brainhack.org/global2020/).

The following features have been added to the package:
* Support for ICA AROMA denoising, and corresponding strategies. Contribution by Hao-Ting Wang htwangtw
* Support for filtering anat compcor noise components based on the type of mask used to calculate the components. Contribution by Steven Meisler smeisler
* Support for scrubbing. The scrubbing components are computed "on-the-fly" based on `mean_fd` and `std_vars`. Contribution by Steven Meisler smeisler
* Better error messages. fMRIprep does not generate all possible confounds by default, so some denoising strategies may not be possible for a particular preprocessed dataset. Users will now get a detailed list of missing noise components for a requested strategy. Contribution by François Paugam FrancoisPgm
* A jupyter notebook with a tutorial on `load_confounds`. Contribution by Michael Weiss m-w-w and Pierre Bellec pbellec
* Refactoring of the code for more modularity and readability. More and better tests. Contribution by François Paugam FrancoisPgm and Pierre Bellec pbellec

Huge thanks to all contributors :tada: :tada: :tada:

Next release will be `beta`, with support for all current fMRIprep types of confounds.

The `load_confounds` 0.6.5 minor release extends support for fmriprep version 20.2.0+, and remains compatible with older versions.
Thanks nuKs for this contribution :tada: :tada:

This is a critical bug fix release.

The derivatives confounds had a NaN in the first row, which caused an error when used in nilearn.
The values of the first row are now duplicates of the second row.
This fix impacts all strategies featuring derivatives expansion of confounds.

Fix contributed by P Bellec, with inputs/reviews from H Shamarke, B Pinsard and A Boré.

This release fixes the pipy package for `load_confounds`.

Both the packaging bug and bug fix were contributed by P Bellec.

This release includes refactoring of the file structure to comply with Pipy standards.

The package is now located in a folder `load_confounds` and `load_confounds.py` was renamed `parser.py`. A similar behaviour to prior module organization can be achieved using:

import load_confounds.parser as lc

All the strategies are being exposed through the package, so it is still possible to run:

from load_confounds import Confounds

or

from load_confounds import Params6


This minor release was prepared by Pierre Bellec.