Configure Grafana Tempo's metrics-generator to produce service-graph and span metrics from ingested traces

Steps

1. In tempo.yaml, add a `metrics_generator:` block with a `processor:` section containing `service_graphs:` and `span_metrics:` sub-blocks to tune each processor (e.g. `service_graphs.histogram_buckets`, `service_graphs.dimensions`, `span_metrics.dimensions`, `span_metrics.intrinsic_dimensions`).
2. Configure `metrics_generator.storage.path` (WAL directory) and `metrics_generator.storage.remote_write` with a list of Prometheus remote_write-compatible endpoints (Prometheus, Grafana Mimir, or Thanos Receive) that generated metrics are pushed to.
3. Set `metrics_generator.registry.collection_interval` (default 15s) to control how often the generator collects and remote-writes its internal metric registry.
4. Explicitly enable the processors per tenant by setting `metrics_generator.processors` (e.g. `service-graphs`, `span-metrics`) in the tenant's `overrides` section — processors are disabled by default even if the `metrics_generator` block is configured.
5. Ensure the distributor/ingest pipeline feeding the metrics-generator is active, since the generator consumes trace data from the same pipeline Tempo uses for ingestion (Kafka-backed in current architecture).
6. In Grafana, add the remote-write target as a Prometheus/Mimir data source and, in the Tempo data source settings, enable the service graph / span-metrics linking to visualize the generated metrics.