Accelerate zenith run splitting option

[gabemery]

The option introduced in PR #19 make use of Observation.select_time on very fine bins. This is very expensive computationally.
Concern line 509 in acceptance_modelisation/base_acceptance_map_creator.py

bloc of code :

compute_observations = Observations()
            for obs in observations:
                time_interval = self._compute_time_intervals_based_on_zenith_bin(obs, zenith_bin)
                for i in range(len(time_interval) - 1):
                    compute_observations.append(obs.select_time(Time([time_interval[i], time_interval[i + 1]])))
        else:
            compute_observations = observations