Read VM metric time series for one instance (Monitoring tab)

Returns chart data for the instance's Monitoring tab: CPU %, memory % (+ page cache, swap), per-disk read/write MB/s, IOPS and latency, per-NIC in/out Mbps, packets, errors and drops. Samples come from the Prometheus that scrapes the compute host's libvirt exporter (30 s scrape, 7 d retention on dev). vCPU steal/delay are deliberately not offered. CPU % (cpu.utilization) is measured on the hypervisor from the vCPU threads' run time divided by the vCPU count; under heavy network/interrupt load it can read higher than the guest operating system shows. Alert rules on cpu.utilization evaluate the same series. A volume's disk metrics appear here, on the instance it is attached to, as the series whose labels.volumeId is that volume's id. Tenant isolation: the caller never sends a query. The backend resolves the instance under the scope and builds every query from a fixed template pinned to that instance's provider server id AND the scope's provider project id; an instance of another project is 404 NOT_FOUND exactly like the detail read. Monitoring switched off: when monitoringEnabled is false no metric is read (and none is billed): 200 with status: MONITORING_DISABLED and an empty metrics array. Turn it on with PATCH .../instances/{instanceId} {monitoringEnabled: true}. An instance that has no provider server yet answers status: NOT_PROVISIONED, also without reading anything. Switched-off spans: Prometheus keeps samples for the retention window whatever the switch says; samples inside the window taken while monitoring was OFF are returned with v: null (a gap), so a customer never sees the span they turned off (and were not billed for). Empty series: status: OK with a metric whose series is empty means the store holds no sample for it in the window (a just-started VM, or memory.utilization on a guest without a balloon driver — Windows needs the virtio balloon driver AND its service). Validation: metric is required, 1–20 names from the closed InstanceMetricName vocabulary; range is 1h|6h|24h|7d (default 1h); step, when given, is 30 seconds to min(range, 86400) seconds and must yield at most 1500 points over the range, otherwise 422 VALIDATION_FAILED. Default steps: 60 s (1h), 120 s (6h), 300 s (24h), 1800 s (7d). Async & state: Synchronous read; no state change. Errors: 401; 403 without compute read access; 404 unknown/other-project instance; 422 invalid parameters; 503 PROVIDER_UNAVAILABLE when the metrics store is unreachable or not configured (never an empty chart that would look like an idle VM); 500 INTERNAL. Idempotency/Concurrency: Safe; no idempotency key.