Some servers like http://geodata.nationaalgeoregister.nl/bag/wfs report wrong layer
extent in their GetCapabilities response.
This commit implements a work around :
- in the 'Request only features intersecting extent' mode, if no feature is returned
in a BBOX enclosing the GetCapabilities extent, then query a single feature to
initialize the extent. The user will then to zoom again on layer and zoom out.
- in the other mode, the extent is updated with the feature geometry extent as soon
as features come from the server, and the user can zoom on layer regularly to se
it updated.
This is related to commit e0d38ba3f912b7ae6b34e66a44e529ec37394458. In case
we have to deal with an inconsitent raster where the nodata value is set to
a value not representable in the data type, ignore it. Otherwise, a NaN value
in a Byte raster would be cast as 0.
QGIS had two problems:
1) It was using the specified legend URL only if its mime type was matching
the layer's mime type. There is no reason for that.
2) When QGIS was using the default layer (empty string), it was not even
trying to find out in what style to pick the legend URL.
When there is a layer group with several sub-layers, the group has a
"default" style and the sub-layers each have a "default" layer. QGIS
was showing two "default" styles for the sub-layers, which would be
confusing to the user and could pick the wrong legend for the
sub-layer if the user picked the wrong entry (the first one).
Had to create a static lib for wmsprovider in order to unittest it.
Button has been changed to "remove", to make it clear that
the button just removes the connection, it does not delete
the database! (It was confusing for spatialite dbs)
Also add tooltips to buttons for extra clarification.
On travis-ci environment, Python version seems to be affected by [this bug](https://bugs.python.org/issue21591).
One way to fix it is to use the old statement instead of exec() function.
