Add more kernels to EventRate

[cboulay]

Currently, EventRate uses a tumbling (non-overlapping) window then sums along the window length.

However, this can be re-imagined as a sparse convolution with a boxcar function.

In a sparse convolution, if the kernel can be represented by a simple function then we need only evaluate the function at each event by extrapolating from the last value (probably the last event) then adding the current event. Finally, for each bin, we extrapolate to the trailing edge to get the instantaneous rate.

This would enable us to add other kernels.
e.g.,

• http://elephant.readthedocs.io/en/latest/reference/kernels.html
• https://github.com/shimazaki/AdaptiveKDE/blob/master/adaptivekde/ssvkernel.py

[cboulay]

Addressed (as a first attempt anyway) in #26