Patching shares mutable values with the input diff

`patch()` inserts mutable values from a diff directly into the destination. Mutating the patched result then changes the diff and affects subsequent applications, even with the default `in_place=False`. This occurs for both added and replaced values.

```python
from dictdiffer import diff, patch
changes = list(diff({}, {"value": {"items": [1]}}))
result = patch(changes, {})
result["value"]["items"].append(2)
assert patch(changes, {}) == {"value": {"items": [1]}}
```

The assertion fails on current master because the second result contains `[1, 2]`. The diff should remain reusable independently of the objects produced by applying it. Regression tests reproduce this for add/change and both values of `in_place`.

Start at the patch() implementation and trace how added and replaced values from the diff are inserted into the destination. Add regression tests for add and change cases with both in_place values, verifying that mutating one patched result does not alter the reusable diff or later results.

Written by the indexing model from the issue text.