Slow /metrics endpoint

[r4victor]

The /metrics endpoint p95 is 2.4s in Sky presumably due to run metrics generated via a ~2s select that scans and joins the entire run table:

dstack/src/dstack/_internal/server/services/prometheus/custom_metrics.py

Lines 91 to 97 in 26247fd

	res = await session.execute(
	select(ProjectModel.name, UserModel.name, RunModel.status, func.count(RunModel.id))
	.join_from(RunModel, ProjectModel)
	.join_from(RunModel, UserModel, RunModel.user_id == UserModel.id)
	.group_by(ProjectModel.name, UserModel.name, RunModel.status)
	.order_by(ProjectModel.name, UserModel.name, RunModel.status)
	)