Create collectd [python] plugin for writing metrics to file

[dpb587]

Currently we're using nc to locally store the metrics collectd is gathering. This is the lazy way. This should really be a collectd plugin of some sort to avoid the extra process and graphite plugin delay.

See also: #4 (comment)

[dpb587]

Note, this can also lead to partial metric names getting into the system because the collectd builtin does buffered writes which leads to incomplete lines in the file which may get picked up when processes are restarted. Can result in confusing terms aggregations on metric name.

$ tail /var/vcap/sys/log/logsearch-shipper/metrics-collectd.metrics.log
host.df_xvda1.df_complex_reserved 170323968.000000 1434730007
host.df_xvda1.df_complex_used 1343905792.000000 1434730007
hos