- Implemented a check in MQTTPusher that prevents the main publish
loop from being invoked anytime mosquitto_loop returns, but only once
every second at most

- Values for MaxInflightMsgNum and MaxQueuedMsgNum are now logged at
startup

- Checking for invalid sensor names in SensorNavigator auto build mode
- More meaningful logging when failing to build sensor trees

- If no sensor hierarchy is specified, SensorNavigator will build a
tree by interpreting each sensor name as a dot-separated path
- Minor changes to README

- MaxMsgNum configuration parameter now defaults to 0 (no message cap)
- The sensor hierarchy string is now logged at startup
- A rebuild of the SensorNavigator is now triggered only at plugin
reloads over the REST API

Make use of mosquitto_max_inflight_messages_set() and the new mosquitto_max_queued_messages_set() functions to limit the amount of memory spent in mosquitto for queued messages.
Rework the handling of disconnects/reconnects

- REST API now correctly performs data analytics computations on
duplicated streaming analyzers
- In duplicated streaming analyzers, the unit that is exposed through
getUnits() is only the one on which that specific analyzer is working.
The units of the other sibling analyzers are still accessible through
the internal _units vector

- When instantiating analyzer units, the sensor tree can be navigated
vertically through the "topdown N" and "bottomup N" keywords, which
respectively mean "N levels down from the top" or "N levels up from the
bottom" of the sensor tree
- This renders the "unit - X" syntax obsolete

Add isTemplate parameter to readSensorBase() and readSensorGroup() to reduce debug output for templates

- Mainly fixes to unsigned-signed comparison warnings

- Small fixes to prevent compilation warnings

- If no analyzerPlugins block is present in global.conf, initialization
of the data analytics framework will be skipped without terminating the
pusher

- Analyzers in on-demand mode now cache units that have been queried
and resolved recently, to speed up response times and reduce overhead
- Once the cache reaches its maximum size, the oldest entry is removed
upon every insertion

- When the analyzer is created, we check whether it only has the root
unit; if so, the unit is immediately resolved and instantiated even if
the analyzer is on-demand

- Auto-publish has been split in two parts: construction of the
sensor names is performed within the Configurators, whereas the MQTT
messages are sent from the MQTTPusher
- This way, sensors always have the correct names upon plugin reloads
- Will make switching to textual MQTT topics easier

- By setting the "streaming" attribute to false, users
can instantiate analyzers that are available only on-demand for requests
on the REST API
- The input/output config will be used to instantiate an unit on the fly
starting from the node queried by the user
- Computation is performed and the result is returned, no data is pushed
on MQTT or to the database

- Added basic obsolescence checks in QueryEngine
- REST API GET requests now output actual JSON when requested
- Other minor changes