Normalization of exponential histograms correctly handles scale changes

[dashpole]

Follow-up from #360 (comment).

For normalization of exponential histograms:

1. If the "start" histogram scale is "lower resolution", then rescale the newest point to the start histogram scale and normalize.
2. If the "start" histogram scale is "higher resolution" then rescale the start point to the new resolution and REPLACE the cache with this rescaled version, using this version from now forward.
3. Always ensure scale value is the "lowest resolution" of start, current.

If OTEL hasn't provided a "rescale" method for ExponentialHistogram, it should.

Side note: The "offset" for buckets shouldn't be as important, but MAY matter if we need to lower resolution even further. The crux of the algorithm should remain the same.

[aabmass]

If OTEL hasn't provided a "rescale" method for ExponentialHistogram, it should.

@dashpole any idea if there is an upstream rescale method yet? Thinking to mark this issue "blocked" until that is available or we could add it upstream. @damemi expressed some interest in working on this.

[dashpole]

marking it blocked SGTM.

[aabmass]

Discussed with team. It would be nice if this was fixed by a processor upstream. We could either log a message or have a self-observability metric to understand if this is happening in practice or not.