Remodnav

Maintenance release:

- Modernized continuous integration setup: Switched from Travis to Appveyor and
GitHub actions, use of a more recent git annex version
- Improved input validation: Especially with lower sampling rates, default or
user-provided parametrizations could lead to inappropriate or impossible
parametrizations for further tooling in remodnav internals. Input validations
for the savitzgy-golay filter and parameters that interact with the sampling
rate have been added to issue warnings to aid users.
- Improved software documentation in the README, contributed by jliebers - thx!
It is now stated explicitly that the input data needs to be tab separated.

- Improve program help

- Ability to distinguish any number of fixation and pursuit events within
a single inter-saccade-period
- No longer use a maximum amplitude parameter to distinguish pursuits from
fixations, but use a single velocity threshold instead. The threshold
is evaluated against a heavily low-pass filtered gaze trajectory, to
only reflect "smooth" eye movement components (and thereby suppress the
impact of measurement noise).
- New parameter `noise_factor` that influences the adaptive saccade velocity
threshold. The saccade onset velocity threshold is the median of all
sub-threshold velocities plus `noise_factor` times the MAD of these
velocities. The saccade peak velocity threshold is computed in the same
fashion, but uses `2x noise_factor`. The default value should work for
noisy data. Reducing this factor can boost saccade detection sensitivity
for clean data (e.g. Nyström et al., 2010 use the equivalent of a factor
of 3.0)

- Initial release.