Adapt analyzer to actually use Spark rather than just collect data from HDFS

[gubser]

Currently the analyzer just iterates over the raw data and applies the analysis in the local python interpreter. Instead it should run it in Spark, the thing i have to think about is what happens if a file is corrupt. Data corruption should not raise an analyzer error, instead it should notify the core that it is not happy with the data. (Also see mami-project/pto-core#20)

Example: download the ipfix file from the measurement server without having stopped QoF beforehand. the ipfix reader will raise an exception because there is no end signature(?) in the file.