Fix JRUBY-6892: Calling #drop on an Enumerator doesn't work properly

When "drop" is called on an Enumerator, the memory locations for the members of the return value were reused.
This resulted in a wrong result that the ticket describes. We address the problem by calling dup() whenever appropriate.

spec/regression/JRUBY-6892_drop_on_enumerator_oddity_spec.rb

@@ -0,0 +1,10 @@
+require 'rspec'
+
+describe "Enumerable#drop" do
+  context "when called on an Enumerator" do
+    let(:enumerator) { (1..10).to_a.each_slice(3)}
+    it "should behave as if it is called on an Enumerable" do
+      enumerator.drop(2).should == [[7,8,9],[10]]
+    end
+  end
+end

src/org/jruby/RubyEnumerable.java

@@ -280,7 +280,10 @@ public IRubyObject call(ThreadContext ctx, IRubyObject[] largs, Block blk) {
                     IRubyObject larg = checkArgs(runtime, largs);
                     synchronized (result) {
                         if (i == 0) {
-                            result.append(larg);
+                            // While iterating over an RubyEnumerator, "arg"
+                            // gets overwritten by the new value, leading to JRUBY-6892.
+                            // So call .dup() whenever appropriate.
+                            result.append(larg.isImmediate() ? larg : larg.dup());
                         } else {
                             --i;
                         }