New way of storing Detections

[Matthew Whiting]

Use run-length encoding to store the detections, rather than storing every single pixel. This will mean major re-writing of parts of the code, but will improve memory management and execution time.

For instance, the fluxes of the individual pixels will not be stored in the detection class -- it will instead be just a map of where the detections are, plus the usual collection of derived parameters such as total flux and the wcs-related parameters.

[Matthew Whiting]

This seems to be working fine now, and has been incorporated into the trunk in revision [258]. Closing this ticket.