refactor: extract backup/restore into config scripts, trim comments

Move the postgres and clickhouse backup/restore hook logic out of inline
compose labels into dedicated pg_backup.sh / clickhouse_backup.sh config
scripts (the pattern other recipes use), and trim the verbose explanatory
comments down to the essential rationale, now living in the scripts.

pg_backup.sh

@@ -0,0 +1,29 @@
+#!/bin/sh
+
+set -e
+
+# The dump lives at the db-data volume root: backup-bot-two v2 snapshots paths inside
+# named volumes (backupbot.backup.volumes.db-data.path), not the container root fs.
+DUMP=/var/lib/postgresql/data/postgres.dump
+
+backup() {
+  pg_dump -U "$POSTGRES_USER" -Fc "$POSTGRES_DB" | gzip > "$DUMP.gz"
+}
+
+backup_cleanup() {
+  rm -f "$DUMP.gz"
+}
+
+restore() {
+  gzip -d "$DUMP.gz"
+  # --if-exists: otherwise DROPs on objects absent from the live db error out and
+  # pg_restore exits 1, killing the chain and leaving the dump behind.
+  pg_restore --clean --if-exists -U "$POSTGRES_USER" --dbname="$POSTGRES_DB" < "$DUMP"
+  rm -f "$DUMP"
+  # pg_restore --clean recreates objects under the live app, so its pooled connections
+  # keep stale type-OID caches ('cache lookup failed for type ...' crash loops, e.g.
+  # Oban). Terminate them so Ecto reconnects fresh.
+  psql -U "$POSTGRES_USER" -d "$POSTGRES_DB" -c "SELECT pg_terminate_backend(pid) FROM pg_stat_activity WHERE datname = current_database() AND pid <> pg_backend_pid();"
+}
+
+"$@"