Slow lazy loading in BlackrockIO

[JuliaSprenger]

Currently lazy loading is implemented by loading the complete dataset in order to determine the size of data objects, in particular AnalogSignals. Although it is faster than loading without lazy loading it could be very much improved by calculating the size based on the binning of the data and the provided t_start and t_stop values.

[samuelgarcia]

In the new rawio and BaseFromRaw, this is done done as follow:

• lazy=False, time_slice=None: the data is read entirely
• lazy=True, time_slice=None: the data is NOT read but lazy_shape give the full shape of Analogsignals
• lazy=False, time_slice=(t1, t2): the data is read only between t1 and t2
• lazy=True , time_slice=(t1, t2): is not allowed, it raise error, this is a non sens to ask for a shape in range you knwon in before.

[JuliaSprenger]

Actually considering that t1 and t2 might need to be adjusted during the loading process (eg. as done in the NeuralynxIO due to gaps in the data) also the last case makes sense and might be useful.