Add slicing on (non-spatial) dimensions at initialization of Arrays

[adjavon]

Goal is to be able to do things slicing through time and/or samples

API example, given a Timelapse dataset of shape T, C, Z, Y, X:

for i in range(10): 
    slices = (i, slice(None), 64, slice(None), slice(None))
    array = open_ds(..., slices=slices) 

Additionally, we can use this momentum to:

• Let it read numpy arrays
• Let it read folders of images
• Use voxel_size instead of resolution in the attributes

[cmalinmayor]

Metadata Conventions:
Current:

• "offset": tuple[int], must be a multiple of voxel size/resolution in each spatial dimension. Represents the offset from the origin in an external coordinate system in real world units
• "resolution": tuple[int]. Represents the size of a voxel in real world units in each spatial dimension
  Desired:
• "units": tuple[str | None], potentially one of a set of enumerated options with known conversion (but potentially just any string). Represents the units that the voxel_size and offset are measured in. Must be small enough that those values are integers. None values represent non-spatial dimensions (e.g. channels).
  -"voxel_size": tuple[int], replaces "resolution". Represents the size of a voxel in real world units in each spatial dimension. Non-spatial dimensions are not included.
• "offset": Optional[tuple[int]], must be a multiple of voxel size in each dimension. Represents the offset from the origin in an external coordinate system in the real world units specified by "units". Defaults to zero in all spatial dimensions if not provided.
• "axis_names": tuple[str], one of a set of enumerated options ("c", "t", "z", "y", "x" is a potential minimal set). Represents the order of the axes of the data that is stored, including non-spatial dimensions.

[cmalinmayor]

Note: We pretty much decided to wrap TensorStore to get our desired functionality (slicing without loading into memory).

[pattonw]

I think everything in here is done now with the exception of opening folders of images. I'll make a separate issue for that.