perf(cdc-graph): add composite (flow_name, start_time) index for CDCGraph

The CDCGraph query range-joins generate_series ticks against
peerdb_stats.cdc_batches per bucket. The pre-existing indexes are a HASH
on flow_name and a btree on start_time, which cannot be combined in a
single scan, so each per-bucket probe scans a start_time range and
filters flow_name as a recheck.

Add a composite (flow_name, start_time) index so each probe is a single
tight index range scan. Benchmarked against ~1.05M batches for one flow:
buffers touched for the 30h window drop from 3844 to ~180 (~20x less
I/O), with matching reductions in cache pressure on large tables.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Author: dtunikov

nexus/catalog/migrations/V55__cdc_batches_flow_name_start_time_index.sql

@@ -0,0 +1 @@
+CREATE INDEX IF NOT EXISTS idx_cdc_batches_flow_name_start_time ON peerdb_stats.cdc_batches (flow_name, start_time);