DOC: sparse.csgraph: Clarify minimum_spanning_tree behaviour in non-connected case. (scipy#17287)

AFAICT, Kruskal's algorithm (as implemented in _min_spanning_tree.pyx)
works as is for the non-connected case as well: the loop
`while i < n_data and n_edges_in_mst < n_verts - 1` simply keeps going
while never finding an edge that connects separate components, and the
separate components are effectively processed independently.

The behavior is also consistent with networkx's minimum_spanning_edges.

Author: anntzer

scipy/sparse/csgraph/_min_spanning_tree.pyx

@@ -50,6 +50,9 @@ def minimum_spanning_tree(csgraph, overwrite=False):
     Small elements < 1E-8 of the dense matrix are rounded to zero.
     All users should input sparse matrices if possible to avoid it.
 
+    If the graph is not connected, this routine returns the minimum spanning
+    forest, i.e. the union of the minimum spanning trees on each connected
+    component.
 
     Examples
     --------