Investigate options for controlling downsampling

[Miles-Garnsey]

At present, when Cassandra produces metrics, they are added to MCAC via a push model. MCAC then writes the metrics out into files for consumption by collectd.

This leaves MCAC with no options to downsample metrics in cluster which have a large number of tables. Ultimately, this can lead to excess resource consumption and further issues downstream (e.g. adding too many metrics to collectd can overwhelm it).

This ticket tracks work to investigate the addition of a configurable throttling mechanism at the MCAC level, so that users can downsample metrics scrapes to avoid overloading collectd.

Some options for configuring such a mechanism are as follows:

1. Set a maximum metrics collected per scrape and adaptively drop samples to ensure it isn't breached.
2. Set a fixed % samples to drop.
3. Set some sort of aggregation mechanism instead of just dropping samples.