0523 — The metric point type is Measurement, not Sample

Context. 0522 named the message type metrics.Sample. Review judged “sample” monitoring-insider vocabulary a first-time user would not recognize, and “Metric” alone misnames the struct — a metric is the named series, the struct is one dated value of it (the CLI’s list shows 36 rows but only 21 distinct metrics). Surveyed: Prometheus “sample”, otel API “measurement”, OTLP/Datadog/Influx/GCM “point”, CloudWatch “metric datum”.

Decision.

• metrics.Sample -> metrics.Measurement (otel’s API word, plainest English: what was measured, the value, when). NewMeasurement, MeasurementKey.
• Name consts become Metric* (MetricUnclaimedWorkers …) with MetricNameReservedPrefix — those constants are metric names, not points.
• Ripple renames keep each surface’s noun honest: admin ListMeasurements / ListMeasurementMessages, otelvulkan RegisterMetricInstruments (one instrument per metric name), CLI list footer counts “series”.

Consequences. User-facing docs never say “sample”; “metric” stays free to mean the named series. Rejected: keeping Sample (user-vetoed), Metric (wrong referent, metrics.Metric stutter), Point/MetricPoint (runner-up, less self-explanatory than Measurement).