Consider to remove IntHistogram and offer only DoubleHistogram (a.k.a Histogram)

[bogdandrutu]

This decision was made in the initial beta release of the metrics, but after using this data model in the collector for some time, I came to the conclusion that they are a bit harmful (I have a bunch of places where I have to deal with conversion for this extra type). By using only the DoubleHistograms we may lose precision for the sum and for the value in the Exemplar (see #256 for details about exemplar). Same argument as described in the Exemplar case can be applied for the sum, and we also can consider the fact that usually when exporting a Histogram users will often look at the percentiles which will be anyway approximations and precision will be lost there regardless of the sum being accurate.

We will keep the IntGuage/IntSum for special cases where people really care about precision for these large numbers.

The deprecation will work this way:

• Rename current DoubleHistogram to just Histogram.
• Deprecate IntDouble from the metric types we support.
• When receiving a deprecated IntHistograms type, the receiver can convert it into the new Histogram, only operation needed is to convert the int64 sum into double. Exemplars will be handled by Consider to remove IntExemplar and offer only DoubleExemplar (a.k.a Exemplar) #256.
• In the collector we will do this in the OTLP receiver, and will remove access to the deprecated field.

/cc @jmacd @open-telemetry/specs-metrics-approvers

[jmacd]

Strong support.

Can you strike this line?

Deprecate IntDouble from the metric types we support.

It seems to conflict with

We will keep the IntGuage/IntSum for special cases where people really care about precision for these large numbers.