Task name handling fixes

Changes task name handling to avoid errors when one task's name is a substring of another task's name in the same session, e.g. "task-rest" and "task-rest2"

Add non-eroded WM / vent masks outputs in MNINonLinear directory

add non-eroded versions of the fMRI-resolution WM and ventricle masks to the outputs in the MNINonLinear directory; this is mainly for NHP data as the default erosion (using a 1-voxel-width Gaussian kernel) yields an empty mask in some cases.

add CLI options --legacy-motion-filter, --no-gsr

Adds two new options:
--legacy-motion-filter (set this to make bandstop motion filter functionality match 4.0.x, i.e. apply filtfilt twice instead of once when run with the default motion_filter_order = 4)
--no-gsr (disables global signal regression)

quick bugfix

Removes leftover debug code from rc0 causing an error outputting filtered movement regressors

Motion filter revision

A minor version update, which makes a change to the default implementation of the bandstop motion filter.

For context, the default filter behavior to date has been to apply a second-order Butterworth filter bidirectionally with filtfilt() twice, meaning that when the default motion_filter_order = 4 was used, the filter was actually 8th-order.

This revision halves the number of motion filter repetitions when using filtfilt(), but makes no change to the unidirectional filter options.

(In testing the revised default filter on resting-state data of a sample of 51 ABCD participants, using filter frequency band of [18.582, 25.726] bpm, post-filtering mean FD changed from 0.1875 to 0.2085 mm, and proportion of frames with FD < 0.2 changed from 75.58% to 72.33%.)

Credit to user @cindyhfls for raising awareness of this issue.

Outlier detection fix / recompile with R2019a

Binaries have been recompiled with Matlab R2019a (instead of R2016b as in previous release); this fixes an issue which caused outlier detection to be omitted (note that this only affected releases compiled with R2016b, not R2017a)

Compile binaries with fix from 4.0.11

The matlab binaries have been recompiled with R2016b to include the zero padding fix from v4.0.11

Add padding to bandpass filter

Added zero padding to the BOLD data prior to applying the bandpass filter to correct artifact that was introduced at the beginning and end of runs. Without padding the filter results in signal deflection at the edge of the timeseries because it requires neighboring time points in order to be applied correctly. After the padding is applied the signal deflection is entirely removed.

Revert Matlab version to R2017a

Changelist from v4.0.9:

FIX: recompiled binaries with Matlab R2017a (v9.2) -- fixes matlabpath error that occurred when compiled with R2019a
MAIN: remove bin directory from gitignore

Update Matlab version and bug fix in filtered_movement_regressors

Fix issue with high TR and matlab version updated to 2019a