Bug in _minimize_local_discrete

[veni-vidi-vici-dormivi]

I mentioned this already in #413 but then forgot about it. Consider the following case in _minimize_local_discrete:
The very first localized covariance matrix is singular and thus returns a negative loglikelyhood of inf.
Then current_min is still inf. res is also inf, a warning gets raised but we continue. res < current_min in this case becomes float("inf") < float("inf") and yields False. Thus we land at sel = i - 1 and sel = -1. We return sequence[-1]`, in this case the last localization radius, without ever checking its likelihood or if the resulting localized covariance matrix is not singular. In my opinion, this is a bug and we should fix it before releasing MESMER v1.0.

It can be fixed by actually skipping results that are inf (rather complicated) or raising an Error if the first element already results is inf if we think that once the localization radii return singular matrices the rest will also always return singular matrices. This is inuitively true when the radii are ordered (which we do not explicitly require) and higher radii lead to less localization and thus to "too little regularized" matrices. At the moment, we always stop when res == inf and return the element from the iteration before (NOTE: I think the comment in the code is misleading).

mesmer/mesmer/core/utils.py

Lines 54 to 80 in bd011cf

	current_min = float("inf")
	# ensure it's a list because we cannot get an item from an iterable
	sequence = list(sequence)
	element = None
	for i, element in enumerate(sequence):
	res = func(element, **kwargs)
	if np.isneginf(res):
	raise ValueError("`fun` returned `-inf`")
	# skip element if inf is returned - not sure about this?
	elif np.isinf(res):
	warnings.warn("`fun` returned `inf`", OptimizeWarning)
	if res < current_min:
	current_min = res
	else:
	# need to return element from the previous iteration
	sel = i - 1
	if sel == 0:
	warnings.warn("First element is local minimum.", OptimizeWarning)
	return sequence[sel]
	warnings.warn("No local minimum found, returning the last element", OptimizeWarning)
	return element

Also check the following test cases:

@pytest.mark.parametrize(
    "values, expected",
    [((5, 4, 3, 2, 3, 0), 3), ((1, 0, 1, 2), 1), ((float("inf"), 1, 2, 3), 1),
     ((3, 2, 1, float("inf"), 0), 4), ((float("inf"), float("inf"), float("inf")), 2),
     ((3, float("inf"), 2, 1, float("inf"), 2), 3)],
)
def test_minimize_local_discrete(values, expected):

    data_dict = {key: value for key, value in enumerate(values)}

    def func(i):
        return data_dict[i]

    result = mesmer.core.utils._minimize_local_discrete(func, data_dict.keys())

    assert result == expected