Release 3.2

[Fabrice Bacchella]

Better management of multiple backend of the same kind. To do that, URI can be used instead of path and the RrdBackendFactory can take a list of active backend that will be checked in order.

Each backend managed and generate URI using different kind of scheme. 

As URI is used internally now, it's dangerous to change the default backend once some rrd are used because it can change the internally generated URI. To avoid that,

either change the default factory once at the beginning or only use URI for rrd's path.

Allows longer that 20 characters in datasource names, the new limit is now 65535 characters.

A offset is encoded in the datasource name, using the private use area from unicode (https://en.wikipedia.org/wiki/Private_Use_Areas). So if a long name is used and the rrd read

by an old library, the name returned will be the long name reduced to about 17 or 18 characters, depending of the length of the file, with a 

few undisplayable characters appended. But if this area is already used for some custom usage, some strange results will happens.

RrdBackend.setFactory should not be public it's not to be used by end user. But to late to make it private and final as it should be.

MongoDB backed end updated to 3.5.0.

Adding an experimental features, that try to downsample before rendering, see http://skemman.is/en/item/view/1946/15343 for rational. It's used

by calling RrdGraphDef.setDownsampler(). It must be given an implementation of DownSampler. Two example one, eu.bengreen.data.utility.LargestTriangleThreeBuckets and

eu.bengreen.data.utility.LargestTriangleThreeBucketsTime are given.

Findbugs run resolve somes bugs.

RrdDb is now java.io.Closeable.

Not yet ready for Java 9.