Allow loading of non-continuous data segments (DataStub.load)

[jens-k]

The method types.untyped.DataStub.load allows addressing a data subset using indices. For example, data.load([23, 2063], [42, 2245]) would give you rows 23 to 42 and columns 2063 to 2245.

However, e.g., when the data represents an LFP, one might be interested in specific non-adjacent rows, i.e. channels. Currently, in order to minimize memory load, one needs to retrieve channel by channel or the first up to the last channel and subselect afterwards. Allowing the user to provide an array with desired channels would feel more natural.

https://nwb-users.slack.com/archives/C5XKC14L9/p1589494856175700

[bendichter]

See https://github.com/NeurodataWithoutBorders/matnwb/issues/120

[lawrence-mbf]

259 fixes this issue