fix: use WHERE filter instead of LIMIT for pgss queryid cardinality control

Replace ORDER BY calls DESC LIMIT 100 with:
  WHERE calls >= 5 OR exec_time_total >= 1000

This keeps all meaningful queries (both frequent AND rare-but-slow) while
filtering PostgREST noise (one-off queries with <5 calls and <1s total time).
The sample_limit: 5000 in prometheus.yml acts as a safety net if cardinality
is still too high.

https://claude.ai/code/session_01SzJxzZNQjDQphaHyaX3RU7

Author: claude

config/pgwatch-prometheus/metrics.yml

@@ -1035,8 +1035,7 @@ metrics:
           temp_bytes_read::int8 as temp_bytes_read,
           temp_bytes_written::int8 as temp_bytes_written
           from aggregated_statements
-          order by calls desc
-          limit 100
+          where calls >= 5 or exec_time_total >= 1000
       17: |
         with aggregated_statements as ( /* pgwatch_generated */
           select
@@ -1083,8 +1082,7 @@ metrics:
           temp_bytes_read::int8 as temp_bytes_read,
           temp_bytes_written::int8 as temp_bytes_written
           from aggregated_statements
-          order by calls desc
-          limit 100
+          where calls >= 5 or exec_time_total >= 1000
     gauges:
       - calls
       - plans_total