otelcol.connector.spanmetrics should initialise stale counter series at 0

[dmurphyccs]

Request

The Span Metrics connector has a metrics_expiration argument which defines the time period after which metrics are considered stale and are removed from the cache.

This is a problem for series that are removed frequently, because traces_spanmetrics_calls_total gets initialised at the value of the first scrape once the series gets registered, and Prometheus skips the jump from null to a given value. In order not to lose that first value, a first value of 0 should be pushed when a new series is added to the counter.

The same problem existed in Tempo metrics-generator but was resolved in this issue: grafana/tempo#2006

Use case

So that infrequent spanmetrics generated by otelcol.connector.spanmetrics in Alloy can be correctly queried in Prometheus, using PromQL functions such as increase() etc.