Normalization by bin volume for overflow bins in binning plugin is broken

[ikbuibui]

If overflow bins are enabled, the option to normalize by bin volume is completely broken if using a log axis, and is subtly broken if using a linear axis.
The core of the issue is that there is no straightforward definition of bin width for the overflow bins, so defining normalization for them is problematic.
One possible fix would be to not normalize the overflow bins, but this would lead to different parts of the histogram data meaning different things. It is not clear to me how this should be tackled.
Current behaviour :

• log axis - It normalizes the first n-2 bins (so including bin 0, the first overflow bin) using the log axis bin widths which are well defined and then does an illegal memory access on device for the final two bins, leading to nans observed by @pordyna
• linear axis - All bins are the same size for a linear axis, and currently the overflow bins are normalized using this constant factor.