Velcolfix branch (for velocity columns memory uninitialization? in linux version 4.2.1197)

[flyingfalling]

This does several things:

1. clarifies the meaning of the various sample columns while they are being read in edf_read.py from the edf_get_float_data() by reading them into a separate Nx1 array linked to the column name key in a dictionary rather than an NxC numpy ndarray. dtypes are the native dtypes returned by the FSAMPLE struct, rather than the all-float version previously returned. One can afterwards run "convert_dtypes" to convert to more robust datatypes if necessary.
2. fixes the number_elements problem (already fixed in main, but necessary to fix again if reading method is changed) by resizing each of the Nx1 arrays to the number of samples read.
3. Adds an "exclude_vel_cols" option in parse() to read_edf, default True, which sets all velocity-related sample columns to NaN. Does not effect events. This is to temporarily address the seeming memory uninitialization/struct offset error (cause uncertain) observed in linux version 4.2.1197 at least. I am not sure how to add "tests" for this in pytest since it requries running multiple processes and is nondeterministic... I added several directories to the tests/ directories which contain python scripts to check this (in R using eyelinkReader, and python using pyedfread)
4. Adds an "ftime" option to read_edf, which converts the "Time" column to floating point (from default uint32) and adds 0.5 seconds for appropriate rows based on flag. I suppose I need to add a test for this, which requires adding an EDF file recorded at 2000 Hz and a way of identifying what is correct (via e.g. ASC from edf2asc). I will prepare this later...