Implement multithreading

[martinghunt]

• run each cluster in parallel
• changed the output of each cluster to write to its own log file, otherwise all output to stdout is a mess when threads>1
• SPAdes can fail from spurious mapping causing clusters with only a few reads. When this happens, write a short message to stderr and point the user to a log file containing the output of spades. The old behaviour was intended for debugging and made users think that something really was wrong.