Support exponential histograms in metric pipelines

[jmacd]

Is your feature request related to a problem? Please describe.

OTLP v0.11 supports an exponential histogram data point. The problem is, many exporters and processors will not initially know what to do with this data point, so some kind of configurable default behavior may be desired. The options are:

1. Reject the new data type if not handled, signal errors
2. Convert the new data type to fixed-boundary explicit histogram data points
3. Convert the new data type to adjustable-boundary explicit histogram data points

Of these, (3) is considered not useful because any destination that supports variable-boundary histograms should accept the new exponential histogram data point quickly.

Describe the solution you'd like
Add some kind of support for metrics pipelines to correctly perform (1) or (2) above out-of-the-box, so that users don't silently drop the new data point. For example, if exponential histograms were converted to a small number of fixed boundaries automatically, users would at least see their data. If exponential histograms become errors that the user is monitoring, that is acceptable too.

Describe alternatives you've considered
Option (3) was the only alternative. The code in this PR open-telemetry/opentelemetry-go-contrib#1353 shows how we could implement option (3), but I doubt anyone wants an auto-scaling explicit boundary conversion automatically.