fix: remove pg_stat_statements from Prometheus sink to prevent cardinality explosion

pg_stat_statements collected via pgwatch-prometheus was creating 30K+ series
in VictoriaMetrics (one per queryid × 16 counter fields), causing query timeouts
and excessive resource usage. This data is already handled by flask-pgss-api
via the Postgres sink, making the Prometheus collection redundant.

Changes:
- Remove pg_stat_statements from the 'full' preset in pgwatch-prometheus metrics
- Reduce pgwatch-prometheus CPU from 1.5 to 0.5 and RAM from 1024m to 512m
  (pg_stat_statements was the main CPU driver)
- Update comments to document the architecture decision

https://claude.ai/code/session_01SzJxzZNQjDQphaHyaX3RU7

Author: claude

config/pgwatch-prometheus/metrics.yml

@@ -2532,7 +2532,8 @@ presets:
       pg_stat_all_tables: 30
       pg_class: 30
       pg_stat_all_indexes: 30
-      pg_stat_statements: 30
+      # pg_stat_statements removed: handled by flask-pgss-api via Postgres sink
+      # Collecting per-queryid metrics here caused 30K+ series cardinality explosion
       pg_stat_replication: 30
       pg_stat_slru: 30
       pg_statio_all_tables: 30

docker-compose.yml

@@ -2,8 +2,9 @@
 # Total allocation: ~5.1 vCPUs, ~7.6 GiB RAM (with headroom for host OS).
 # CPU overcommitment is intentional - containers rarely hit limits simultaneously.
 #
-# pgwatch-prometheus gets higher CPU (1.5) to handle pg_stat_statements load,
-# which was observed spiking to 150%+ under heavy query workloads.
+# pgwatch-prometheus gets higher CPU (1.5) to handle metric collection load.
+# Note: pg_stat_statements is NOT collected here (handled by flask-pgss-api
+# via the Postgres sink) to avoid cardinality explosion in VictoriaMetrics.
 
 services:
   # Config Initializer - Copies static configs from image to volume
@@ -139,8 +140,8 @@ services:
     build:
       context: ./pgwatch
     container_name: pgwatch-prometheus
-    cpus: 1.5
-    mem_limit: 1024m
+    cpus: 0.5
+    mem_limit: 512m
     command:
       [
         "--sources=/postgres_ai_configs/pgwatch-prometheus/sources.yml",