+ when setting channels: raise a Warning if no valid channels were found or are not availalbe

Also, avoid spreading mpi code around the code tree. This makes maniteinance
easier

This helps to not screw up the scanner manager (processed.dat) if *preloaded* chunks could not be read (e.g., if temporarily not mounted, etc.)

Add comment to the prior import of MPI in xcorr.py & setupclass.py:
on EULER, this leads to a fork() warning:
http://gfs.sourceforge.net/wiki/index.php/System(),_popen(),_fork()_and_OpenMPI

+ minor changes

--> logs only the config setup of the master process

+ no "Found XXX detections" for normal run (only for "--verbose")

- more arguments /options
- better logging & comments

- optional: when plotting "grouped" clusters: sort by cluster-size
- optional: split clusters with higher thresholds
- reworked cluster ID (now is string)
- for maxclust criterion: approximate clustering-threshold
   (to build dendrograms that make sense)
- optional: reprocessing of events waveforms (incl. equalizing)
- new for generating templateset:
    - when a cluster has len == 2 --> take the smaller event
    - optional: can output reprocessed template waveforms
- misc fixes & improvements

(because Toni unnecessarily changed it before) ;-)

1. don't FFT_filter for procType='highpass'
2. _waveform_processing_trace()
    - reorder args
    - don't do anything if procType='raw' (return trace right away)

changed config.stop50Hz to config.noiseStops
    --> can have more than one frequency to notch away

+ improve checking frequency resolution (now independent of bandstopped frequency and fs)
   --> necessary length for a *stable* notch filter: ~25s