memory issues with prof_class()

[TillF]

prof_class() runs out of memory with large files of rstats.txt.
This has been partially reported in #16

[TillF]

Significantly improved with branch large_prof_class (#30 ), but there are still some redundancies that could be eliminated (marked in the code with ii).
Added some suggestions to the user in the FAQ (wiki).

[TillF]

We could also exploit the rstats_head.txt better: currently, setting 0 or 1 for an attribute means the same: Do not make any clustering on it.
It would be nice to have
1: do not make any clustering on this attribute, but keep it in the output.
0: completely ignore this attribute. This would allow removing attributes even after area2catena() if it runs out that they consume to much memory.

[TillF]

The modified version of prof_class of branch large_prof_class does not yet correctly treat the argument EHA_SUBSET.