Reduce perf-lotsofthreads iterations 10x in Debug

The benchmark's purpose is to stress the cross-thread free path
under contention, which is already observable at much lower
iteration counts. Debug builds carry full instrumentation and run
~10x slower per iteration, so 200k iterations across 8 threads
makes this single test dominate Debug ctest wall-time without
producing any additional contention coverage relative to a smaller
count.

Divide iterations by 10 when NDEBUG is not defined. Release builds
keep the original counts (200000 / 50000 depending on platform and
sanitizer) so the benchmark's signal is unchanged where it matters.

Local measurement on Debug:
  perf-lotsofthreads-fast:  ~136s -> 13.6s
  perf-lotsofthreads-check: ~500s -> 49.7s

The test is on the concurrency-stress exclusion list and so is
deliberately not affected by the --smoke knob; reducing iterations
in Debug here is the right lever.

Author: mjp41

src/test/perf/lotsofthreads/lotsofthread.cc

@@ -101,6 +101,14 @@ int main()
 #else
   size_t iterations = 200000;
 #endif
+#ifndef NDEBUG
+  // Debug builds run with full instrumentation enabled and are
+  // ~10x slower per iteration. The cross-thread batch behaviour
+  // this benchmark stresses is observable at much lower counts;
+  // reduce iterations so this test does not dominate Debug ctest
+  // wall-time. Release builds are unaffected.
+  iterations /= 10;
+#endif
 
   int threadcount = 8;
   vector<thread> threads;