Create a mimir tenant for high cardinality postgres metrics
It would be really useful to expose some very high cardinality metrics from postgres to our observability system.
Specifically, for each of the 5000 queries in pg_stat_statements, for each of the roughly 20 database instances we run, it would be really useful to expose roughly 20 metrics (and I'm sure if we were unconcerned about cardinality we would want to add a dozen or so more). So a total of 5000 * 20 * 20 = 2 million
metrics, with possible ~50% growth over time.
Currently we expose only the top 10% of pg_stat_statements queries by runtime, cutting this number by a factor of 10 in current production. This makes the metrics miss some things, so it would be really nice to capture all 5000 queries.
A dedicated mimir tenant for postgres-related metrics would help support this.