feat: Allowing to use zcollection without any dask cluster.

[Thomas-Z]

The objective of these changes is to enable the use of a zcollection without requiring a Dask cluster.

This functionality is made possible by the new distributed parameter, which has been added to all functions and methods that rely on Dask (except for the map and map_overlap functions, as they are specifically designed for Dask usage).

However, the choice of the distributed keyword might not be ideal, given that we are working in a context where the distributed package is heavily used, which could lead to some confusion.
The keyword was already in use within a sub-function, so I adopted it, but this is open for discussion.

@robin-cls: Could you review this and confirm whether it meets your needs?
@fbriol : Could you evaluate the implementation and either approve it or suggest an alternative approach?

[Thomas-Z]

A second commit includes changes that allow loading a specific set of partitions using the new selected_partitions parameter of the collection.load() and view.load() functions.

[robin-cls]

Hello @Thomas-Z thank you for this first implementation. My use case is to use the insert() method in a dask worker without submitting any graph to the current dask scheduler. For now, I am quite busy on another pressing matter, but I think I will be able to test your version by the end of the year if that's not too late for you