Inconsistent auto generated dimension names in numpy_to_data_array

[lucianopaz]

Describe the bug
This was originally reported in pymc-devs/pymc#7572. The problem is that arviz.data.numpy_to_data_array does not yield consistent dimension names when there are default_dims and dims that also match some of these.

To Reproduce

import numpy as np
from arviz.data import numpy_to_data_array

generated_dims = []
for dims in [None, ["chain", "draw"], ["chain", "draw", None]]:
    out_dims = numpy_to_data_array(
        np.empty((4, 500, 7)),
        var_name="a",
        dims=dims,
        default_dims=["chain", "draw"]
    ).dims
    generated_dims.append(out_dims)

assert all(dims == generated_dims[0] for dims in generated_dims)

Specifically:

>>> generated_dims
[('chain', 'draw', 'a_dim_0'),
 ('chain', 'draw', 'a_dim_2'),
 ('chain', 'draw', 'a_dim_2')]

Expected behavior
The result that one gets when dims = None should be the same as when the chain and draw dimension are also specified.

Additional context
Current arviz main