WIP - Demonstrate graph-based metrics

[djaglowski]

This is very incomplete but I want to show the basic idea of adding incoming/outgoing metrics at the graph level.

Obviously this code is pretty gnarly to begin with, and this only makes it worse, but this is superior to other approaches because:

1. All attributes described in Attributes for component instancing #11179 are already available exactly where they are needed.
2. Unlike Add 'pipeline' attribute to processor metrics #11171, changes are not needed elsewhere.
3. Unlike Add ingoing and outgoing counts to processorhelper #10910, these metrics will apply to all processors, not just those built with processorhelper. This benefit translates to all component kinds too though.
4. Additional metrics, such as proposed in [processorhelper] Add throughput metrics #11147 can be added quite easily.
5. For processors and connectors, it's likely we could move span capture to this level.
6. Future functionality such as proposed in Remotetap extension concept #10963 may be inserted in these consumer wrappers as well.

[mx-psi]

I think this is the right approach to do this; it gives us more flexibility in terms of the API and we have the visibility we need at this level. I support this.

cc @open-telemetry/collector-maintainers PTAL so we can discuss the high-level approach before moving forward