Add a tolerance when computing "good segments" for ExposureBins and TimeEnergyBins

[AdamGoldstein-USRA]

ExposureBins and TimeEnergyBins calculate "good segments," which are contiguous segments of data separated by bad time intervals (typically no data). This is determined by comparing the high edge of a bin to the low edge of the next bin, and if they are not the same, then the end of the good segment is declared, and a new segment starts. The implementation in ExposureBins, for example, is

mask = (self.lo_edges[1:] != self.hi_edges[:-1])

There are cases when self.lo_edges[1:] and self.hi_edges[:-1] are slightly different, usually by rounding/numerical precision limitations, especially on older files that use 32-bit floats. This causes an incorrect segmentation of the data. An improvement is to implement a tolerance:

mask = np.abs(self.lo_edges[1:] - self.hi_edges[:-1]) > tol

where tol is the tolerance in seconds allowed. As an example, tol=1e-4 is sufficient for BATSE continuous data.

The keyword option of tol could either be added as an option to the _calculate_good_segments() method and passed via an option in the initializer, or have the initializer set a private tolerance variable and _calculate_good_segments() uses that private variable.

[BillCleveland-USRA]

Move the implementation fro CGRO to Core.

[AdamGoldstein-USRA]

This should also be applied to TimeChannelBins. An example of how this is addressed in gdt-cgro.