Solves #408 (incompatibility between crop and downsampling)

[adehecq]

Resolves #408.

Phew, this was a tough one! It turns out that Raster.crop() was not very thoroughly implemented in the case when data is not loaded. A few things to pay attention to:

• one needs to create a Window using the on-disk transform, not the in-memory one, otherwise we're just reading the wrong bits!
• calling rio.DatasetReader.read is actually more complex than it seems when either of cropping or downsampling need to be executed. One needs to make sure that Window and out_shape really match. The window shape can actually be misleading because with option boundless=False (default), the window will be cropped if it extends beyond the raster extent. It makes sense after all...

Doing so, I noticed that our downsampling functionality is actually slightly wrong, see issue #447. Basically, the sampling interval is not exactly constant.
Because of that, one does not get exactly the same result if cropping a downsampled raster with or without data loaded (see commented test). This should be fixed together with #447.

Decided to also modify Raster.info() to print to screen by default, as mentioned in #361.

[rhugonnet]

Amazing! No comment on the code!

My only question: While you are still in this, what do you think is the best way to fix the downsampling issue consistently? #447

[adehecq]

My only question: While you are still in this, what do you think is the best way to fix the downsampling issue consistently? #447

I made some suggestions in the issue directly. I just don't want to have a look at this now, because I'm sure it will take a few hours to get everything sorted and thoroughly tested...