Support custom monitored resources for Metrics

[daniel-sanche]

The exporter currently uses a call to gcp_resource_detector.get_monitored_resource to associate a monitored resource with each metric. But it appears that gcp_resource_detector only supports a very limited set of resources, and otherwise falls back to generic_node. We should have a way of explicitly specifying any other monitored resource supported by GCP

[aabmass]

@daniel-sanche can you add some more context on why this is needed (feel free to reach out internally). Cloud Monitoring only supports a subset of these monitored resources for custom (user) metrics https://cloud.google.com/monitoring/custom-metrics#which-resource

[daniel-sanche]

I was hoping to use this exporter to send built-in metrics for the bigtbale client, but bigtable uses its own custom monitored_resource. It sounds like the bigtable team had to create a custom exporter for Java, but I was hoping to avoid that for python if we can get away with it.

I opened a PR that addresses this by letting users pass in a monitored resource at init, which would have other performance benefits to users outside this niche use-case. But I can also close this and just subclass the exporter in the bigtable client if you'd prefer

[dashpole]

Sorry this got dropped. If we implement this, it should follow the pattern set by GoogleCloudPlatform/opentelemetry-operations-go#854 and GoogleCloudPlatform/opentelemetry-operations-java#346