[ruby/psych] Optimize cache with compare_by_identity

Using `compare_by_identity` gives a 4x performance boost on cache hits.
Benchmark in fastruby/fast-ruby#189

lib/psych/visitors/visitor.rb

@@ -12,7 +12,7 @@ def accept target
       def self.dispatch_cache
         Hash.new do |hash, klass|
           hash[klass] = :"visit_#{klass.name.gsub('::', '_')}"
-        end
+        end.compare_by_identity
       end
 
       if defined?(Ractor)

lib/psych/visitors/yaml_tree.rb

@@ -80,7 +80,7 @@ def initialize emitter, ss, options
           raise(TypeError, "Can't dump #{target.class}") unless method
 
           h[klass] = method
-        end
+        end.compare_by_identity
       end
 
       def start encoding = Nodes::Stream::UTF8

test/psych/test_ractor.rb

@@ -24,6 +24,9 @@ def test_not_shareable
   end
 
   def test_ractor_config
+    # Config is ractor-local
+    # Test is to make sure it works, even though usage is probably very low.
+    # The methods are not documented and might be deprecated one day
     assert_ractor(<<~RUBY, require_relative: 'helper')
       r = Ractor.new do
         Psych.add_builtin_type 'omap' do |type, val|