Periodic telemetry API improvement

[clhain]

Related to #1302

Currently, receiver/processor/exporter implementations are required to remember to call effect_handler.start_periodic_telemetry(), and if not done, there are no metrics surfaced until the pipeline is shutdown. This is something that can/should probably be handled at a lower level in the engine.

This issue is to investigate how to handle this correctly.

The API will likely need to support use-cases like perf exporter where the collection interval is specified as part of the node's config:
https://github.com/open-telemetry/otel-arrow/blob/main/rust/otap-dataflow/crates/otap/src/perf_exporter/exporter.rs#L135