Pixel-clusterizer

Bug fixed:
- No crash if hit array is empty

Enhancements:
- Example how to add col/row distance calculation added
- Hit field definition simplified and checks added
- End of cluster/event functions can be jitted manually to allow for special jut arguments
- Small code clean up

- Noisy pixel mask supported
- Disabled pixel mask supported
- Min/max hit charge can be set
- API changes (e.g. array size does not have to be specified)
- More tests / coverage
- Clean up

This release changes the cluster data format! The mean column / row are not centered anymore.

Consider a cluster with on pixel at column / row = 1 / 2

Before:
cluster: mean column / mean row = 1.5 / 2.5

Now:
cluster: mean column / mean row = 1.0 / 2.0

This adresses issue 2 and solves the discrepancy that cluster seeds were not shiftet by 0.5.

The is a feature release with the following changes:
- Numba 0.24 is supported now and thus structured arrays. All record arrays were changed to structured arrays
- A pure python mode was added that is useful for testing. Coverage testing and reporting were added using the pure python mode.

- The data type and field names of the cluster array can be changed freely
- BUG: all hit data is now set in the clustered hit array

Feature Release:
- The clusterizer works now on arbitrary hit arrays (field names and types can be chosen freely)
- End-of-event and end-of-cluster function can be defined by the user to do custom calculations
- The cluster output can be extended by additional field