Remove the gather_index in DataAssembly

[YingtianDt]

Hi, I suggest removing gather_index in any DataAssembly types due to the following reasons:

1. it always create bugs when one dimension only has one coordinate. This creates big confusion for new users;
2. it has no practical use;
3. occasionally, multi-index creates more trouble than the normal index: e.g., when we need to add more coordinates.

Though this issue is in brainio repo, I am raising it here since it is more relevant in context.

[mike-ferguson]

Hi @YingtianDt - thanks for the suggestion and for opening an issue! I will pass this forward to the rest of the team, and we appreciate the contribution.

[mschrimpf]

related: brain-score/brainio#42

[YingtianDt]

Hi, I am moving the discussion of this to this PR, since me and Martin seem to agree on rather removing gather_index but fixing it (potentially with this PR).

With this said, please still be aware of the downsides of using Multi-index. E.g.,

Take a concrete example here:

a = NeuroidAssembly(
    [[0, 1, 2], [3, 4, 5]],
    dims = ('presentation', 'neuroid'),
    coords = {
        'stimulus_id': ('presentation', ['a', 'b']),
        'stimulus_type': ('presentation', ['m', 'n']),
        'neuroid_id': ('neuroid', ['x', 'y', 'z']),
        'neuroid_layer': ('neuroid', ['k', 'f', 'c']),
    }
)

1. when adding new coordinates, we have to always use gather_index again to maintain the multi-index. e.g.,

a = a.assign_coords(neuroid_weight = ('neuroid', [0.1, 0.2, 0.3]))
a = NeuroidAssembly(a)

2. when comparing two multi-indices, the ordering of the coordinates within each index matters. e.g.,

b = NeuroidAssembly(
    [[0, 1, 2], [3, 4, 5]],
    dims = ('presentation', 'neuroid'),
    coords = {
        'stimulus_type': ('presentation', ['m', 'n']),
        'stimulus_id': ('presentation', ['a', 'b']),
        'neuroid_id': ('neuroid', ['x', 'y', 'z']),
        'neuroid_layer': ('neuroid', ['k', 'f', 'c']),
    }
) # here the ordering of coordinates of presentation is reversed

print(a.presentation == a.presentation)  # this will work
print(a.presentation == b.presentation)  # this gives empty comparision like the following

# array([], dtype=bool)
# Coordinates:
#   * presentation          (presentation) MultiIndex
#   - presentation_level_0  (presentation) object 
#   - presentation_level_1  (presentation) object