Enable test_mutable_ref_output_is_view and add .base verification

Enabled the previously skipped test for mutable reference returns. The implementation already supports this case using PyArray_SimpleNewFromData with explicit .base setting to keep the owner alive.

Added verification that:
- .base is the owner object (not None)
- Array is writable
- Modifications to the array affect the C++ data (true zero-copy view)

This test verifies that non-const reference returns create writable views with proper lifetime management.

Addresses comment 3679483484

Co-authored-by: jpfeuffer <8102638+jpfeuffer@users.noreply.github.com>

Author: Copilot

tests/test_numpy_vector_converter.py

@@ -78,7 +78,6 @@ def test_const_ref_output_is_readonly_view(self, numpy_vector_module):
         assert result.base is not None
         assert result.base is t, f"Expected .base to be the owner object, got {type(result.base).__name__}"
 
-    @pytest.mark.skip(reason="Mutable ref views require ensuring C++ object lifetime exceeds view lifetime - needs investigation of reference handling")
     def test_mutable_ref_output_is_view(self, numpy_vector_module):
         """Non-const ref should create a writable view (zero-copy)."""
         import numpy as np
@@ -93,6 +92,10 @@ def test_mutable_ref_output_is_view(self, numpy_vector_module):
         # Array should be writable
         assert result.flags.writeable
 
+        # Check base attribute - should be the C++ object (self) to keep it alive
+        assert result.base is not None
+        assert result.base is t, f"Expected .base to be the owner object, got {type(result.base).__name__}"
+        
         # Modify array - SHOULD affect C++ data since it's a view
         result[0] = 999.0
         result2 = t.getMutableRefVector()