Tensor

- Fix many issues with timeout handling
- Add Nginx stats sources and a log following source
- Fixes aggregation bug with Munin source
- Updates a pile of documentation and other small fixes

- CPU source now returns individual metrics instead of just a total and supports more kernels
- Adds SNMP source (Requires PySNMP4)
- Adds sFlow source (Requires Construct)
- Adds a network interface stats source
- Adds aggregation support to the event core rather than having each source manage caches and aggregation on their own
- Critical and Warning thresholds now support a regular expression which is matched in the event core after aggregation making it more intuitive to do things like match usage on any disk.
- No longer raise a critical event when a source encounters an unhandled error as this caused issues which could not be cleared automatically, just logs an error instead.

Fixes reconnect issue

- Add a Riak source (contributed by jerith)
- Add a RabbitMQ source
- Fixes re-connection issues
- Fixes some null metrics handling issues and alerting on failed sources.

- Fixes the default package configuration
- Inherit error handling in Source objects to prevent unhandled exceptions blocking the check loop
- Fixes silly TCP re-connection defaults exponentially growing to over an hour
- Various fixes and changes to queue handling and TCP back-pressure support
- Added User-Agent configuration to HTTP check

- Migrates to a modular backend for event outputs
- Change how Sources are constructed to create a Tensor queue monitoring source
- Adds support for including a 'conf.d' style directory into the YAML configuration