more_splits_than_threads.test: 64 splits, not 1000

The claim is that splits outnumber readers, so each reader re-initializes a
reused connection several times. 64 splits over 8 threads is eight claims per
reader; 1000 was ~2000 HTTP round trips per query and the slowest file in the
HTTP lanes (10-14 s alone, timing out at 60 s under 8-way load). Expected
results are unchanged: they depend on n, not on the split count.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

Author: rustyconover

test/sql/integration/splits/more_splits_than_threads.test

@@ -1,15 +1,19 @@
 # name: test/sql/integration/splits/more_splits_than_threads.test
-# description: 1000 splits over 8 threads: no duplicates, no drops
+# description: 64 splits over 8 threads: no duplicates, no drops
 # group: [vgi_integration_splits]
 
-# Splits outnumber readers, so each reader claims many in sequence. That is the
+# Splits outnumber readers, so each reader claims several in sequence. That is the
 # shape that exercises re-initializing a REUSED connection: the previous split's
 # stream has to be closed (IPC EOS on stdin) before the next init is written, or
 # the init request lands inside an unterminated stream and the transport mis-frames.
 #
 # COUNT(DISTINCT) is the assertion that matters, not COUNT(*). A drain loop that
 # re-reads a split would still produce 5000 rows if it also dropped one; only
 # distinctness separates "read everything once" from "read something twice".
+#
+# 64 splits is eight per reader at threads=8: every reader re-initializes its
+# connection repeatedly, which is the whole claim. It used to be 1000, which
+# costs ~2000 HTTP round trips per query and proves nothing more.
 
 require-env VGI_TEST_WORKER
 
@@ -25,16 +29,16 @@ SET threads=8;
 
 query III
 SELECT count(*), count(DISTINCT n), sum(n)
-FROM example.main.split_many(n := 5000, splits := 1000);
+FROM example.main.split_many(n := 5000, splits := 64);
 ----
 5000	5000	12497500
 
-# Same claim set, single-threaded: one reader must drain all 1000 sequentially.
+# Same claim set, single-threaded: one reader must drain all 64 sequentially.
 statement ok
 SET threads=1;
 
 query III
 SELECT count(*), count(DISTINCT n), sum(n)
-FROM example.main.split_many(n := 5000, splits := 1000);
+FROM example.main.split_many(n := 5000, splits := 64);
 ----
 5000	5000	12497500