Fix alert rules and prometheus config for stackdriver metrics

[shlbatra]

Summary

• Updated all alert rules in alert_rules.yml to use stackdriver_* metric names (replaced old fastapi_fastapi_* and dataflow_job_user_counter references that no longer exist after the Cloud Monitoring migration)
• Removed otel-collector scrape target from prometheus.yml (OTel Collector is behind local-dev profile, was showing as DOWN)

Alert Rule Changes

Old Alert	New Alert	Metric Source
HighPredictionErrorRate (fastapi_fastapi_*)	HighCloudRunErrorRate	Cloud Run 5xx rate
NoPredictionsFlowing (fastapi_fastapi_*)	NoPredictionsFlowing	Dataflow prediction_success
HighPredictionLatency (fastapi_fastapi_*)	HighPredictionLatency	Dataflow prediction_latency_ms_mean
FeatureFetchFailureSpike (dataflow_job_user_counter)	FeatureFetchFailureSpike	fetch_failure + fetch_retry
HighParseErrorRate (dataflow_job_user_counter)	ParseToFetchDropoff	parse vs fetch success ratio
PredictionBatchFailureSpike (dataflow_job_user_counter)	PredictionRetrySpike	prediction_retry

Cost-attribution alerts were already using correct stackdriver_* names — unchanged.

Test plan

• Verify alert rules load: http://localhost:9090/alerts shows all rules as inactive
• Verify no DOWN targets: http://localhost:9090/targets shows only stackdriver-exporter
• Verify Alertmanager connectivity: http://localhost:9093 reachable

🤖 Generated with Claude Code