Please address the comments from this code review:
## Individual Comments

### Comment 1
<location> `src/strawberry_sqlalchemy_mapper/mapper.py:568` </location>
<code_context>
+                edges.append(
+                    edge_type.resolve_edge(
+                        related_object,
+                        cursor=max(0, decoded_before_index - total_count + i),
+                    )
+                )
</code_context>

<issue_to_address>
Cursor calculation for before pagination may be incorrect.

Please review the logic for calculating the cursor in the 'before' branch, as it may produce incorrect values when total_count is less than decoded_before_index. Ensure the cursor calculation is accurate for all cases.
</issue_to_address>

### Comment 2
<location> `src/strawberry_sqlalchemy_mapper/mapper.py:576` </location>
<code_context>
+                edges.append(
+                    edge_type.resolve_edge(
+                        related_object,
+                        cursor=max(0, total_count - last),
+                    )
+                )
</code_context>

<issue_to_address>
Cursor value for 'last' pagination is constant for all edges.

Using the same cursor for all edges can break pagination. Each edge should have a unique cursor reflecting its position.
</issue_to_address>

Please address the comments from this code review:
## Overall Comments
- The pagination logic in loader_for is quite dense—consider extracting the offset/limit calculation and argument validation into standalone helper functions to improve readability and testability.
- There’s duplicated code for resolving `info.context['sqlalchemy_loader']` versus `info.context.sqlalchemy_loader`; you could introduce a small accessor/helper to DRY this lookup and reduce branching.
- The `_resolve_connection_edges` method has grown large with pagination logic—extracting pageInfo and edge slicing into a dedicated paginator class or module would make both the mapper and loader code easier to follow.

## Individual Comments

### Comment 1
<location> `src/strawberry_sqlalchemy_mapper/loader.py:133` </location>
<code_context>
     """

-    _loaders: Dict[RelationshipProperty, DataLoader]
+    _loaders: Dict[RelationshipProperty, PaginatedLoader] = {}

     def __init__(
</code_context>

<issue_to_address>
Class-level _loaders dictionary may cause cross-instance data leakage.

Since all instances share the same _loaders cache, this can cause unintended data sharing. Recommend moving _loaders to an instance variable to avoid this issue.
</issue_to_address>

### Comment 2
<location> `tests/test_loader.py:198` </location>
<code_context>
+                Employee(name="e4"),
+            ],
+        )
+        d_empty = Department(name="empty")
+        session.add(d1)
+        session.add(d_empty)
+        session.flush()
+        session.commit()
</code_context>

<issue_to_address>
Consider adding assertions for empty result sets.

Please add assertions to verify that loading employees for d_empty returns an empty list, ensuring correct loader behavior for departments without employees.
</issue_to_address>

### Comment 3
<location> `RELEASE.md:119` </location>
<code_context>
+}
+```
+
+Get all books before a spcific book's cursor:
+
+```gql
</code_context>

<issue_to_address>
Typo: 'spcific' should be 'specific'.
</issue_to_address>

<suggested_fix>
<<<<<<< SEARCH
Get all books before a spcific book's cursor:
=======
Get all books before a specific book's cursor:
>>>>>>> REPLACE

</suggested_fix>

### Comment 4
<location> `RELEASE.md:144` </location>
<code_context>
+}
+```
+
+Get the last three books for a specific author before a spcific book's cursor:
+
+```gql
</code_context>

<issue_to_address>
Typo: 'spcific' should be 'specific'.
</issue_to_address>

<suggested_fix>
<<<<<<< SEARCH
Get the last three books for a specific author before a spcific book's cursor:
=======
Get the last three books for a specific author before a specific book's cursor:
>>>>>>> REPLACE

</suggested_fix>