Deduplicate during _flatten_markers()

The potential downside of the approach in this series of commits is that
cnf() and dnf() can, in the worst case, be exponentially expensive.

eg `cnf((a1 and b1) or (a2 and b2) or ... (an and bn))` generates 2^n
intersections.

In practice, markers on python packages seem not to get long enough for
this to be an issue.

Deduplicating the markers on unions and intersections is the lowest of
low-hanging fruit in mitigating this.

src/poetry/core/version/markers.py

@@ -370,11 +370,14 @@ def _flatten_markers(
 
     for marker in markers:
         if isinstance(marker, flatten_class):
-            flattened += _flatten_markers(
+            for _marker in _flatten_markers(
                 marker.markers,  # type: ignore[attr-defined]
                 flatten_class,
-            )
-        else:
+            ):
+                if _marker not in flattened:
+                    flattened.append(_marker)
+
+        elif marker not in flattened:
             flattened.append(marker)
 
     return flattened