Fix type_name_lookup so top-level classes win over nested namesakes

Skipping the ambiguity check when `cm.name == cm.full_name` let a nested
class (e.g. `Gem::SafeMarshal::Elements::String`) register itself as the
unqualified `String` in `unqualified_names`, which then overwrote the
core `String` entry during the final `lookup.merge!`. Remove the skip so
top-level classes participate in ambiguity detection — the second
sighting now correctly marks the name ambiguous and deletes it from
`unqualified_names`, leaving `lookup["String"]` pointing at the core
class's path.

Author: st0012

lib/rdoc/store.rb

@@ -358,7 +358,6 @@ def type_name_lookup
       all_classes_and_modules.each do |cm|
         lookup[cm.full_name] = cm.path
         unqualified_name = cm.name
-        next if unqualified_name == cm.full_name
 
         if ambiguous_names[unqualified_name]
           # already known ambiguous, skip

test/rdoc/rdoc_store_test.rb

@@ -1352,6 +1352,23 @@ def test_type_name_lookup_ambiguous_unqualified_name_excluded
     refute lookup.key?('SubClass')
   end
 
+  def test_type_name_lookup_top_level_class_wins_over_nested_namesake
+    top_level_string = @top_level.add_class RDoc::NormalClass, 'String'
+    top_level_string.record_location @top_level
+
+    nested_string = @top_level.add_class RDoc::NormalClass, 'Gem::Elements::String'
+    nested_string.record_location @top_level
+
+    @s.complete :public
+
+    lookup = @s.type_name_lookup
+
+    # Top-level class retains its own path; nested namesake does not
+    # hijack the unqualified name.
+    assert_equal top_level_string.path, lookup['String']
+    assert_equal nested_string.path, lookup['Gem::Elements::String']
+  end
+
 
   def test_merge_rbs_signatures_does_not_overwrite_inline_annotations
     m = RDoc::AnyMethod.new(nil, 'greet')