Force a major GC and compact before measuring RSS

This will hopefully remove any variance caused by one benchmark run
having just hit a GC and one not.

Author: eightbitraptor

harness/harness-common.rb

@@ -146,7 +146,10 @@ def return_results(warmup_iterations, bench_iterations)
   yjit_stats = RubyVM::YJIT.runtime_stats if defined?(RubyVM::YJIT.enabled?) && RubyVM::YJIT.enabled?
   zjit_stats = RubyVM::ZJIT.stats if defined?(RubyVM::ZJIT.enabled?) && RubyVM::ZJIT.enabled?
 
-  # Collect our own peak mem usage as soon as reasonable after finishing the last iteration.
+  # Full GC then compact before measuring RSS so fragmentation doesn't inflate the number.
+  GC.start(full_mark: true, immediate_sweep: true)
+  GC.compact if GC.respond_to?(:compact)
+
   rss = get_rss
   ruby_bench_results["rss"] = rss
   if maxrss = get_maxrss