Implementation and documentation of max_lt/mix_gt is out of sync or contradictory

The documentation of max_lt and min_gt imply a different behavior ("None is returned if seq was empty or all items in seq were >= val." see https://gitlab.aicrowd.com/flatland/baselines/blob/master/utils/observation_utils.py).

For max_lt I would expect the following to pass, but it does not:

        >>> # Failure cases:
        >>> # None is not returned
        >>> print(max_lt([], 0.5))
        None
        >>> # None is not returned
        >>> print(max_lt([1.0], 0.5))
        None
        >>> # negative values in seq are ignored
        >>> max_lt([-1.0], 0)
        -1.0

min_gt has the same failure cases.

The implementation of both functions could be simpler and more robust. Something like this:

def max_lt(seq, val):
    try:
        return max(x for x in seq if x < val)
    except ValueError:
        return None

def min_gt(seq, val):
    try:
        return min(x for x in seq if x > val)
    except ValueError:
        return None