cca etc: get factor centroids from LC scores instead of WA

In principle LC and WA scores give same centroids, but they differ
numerically. With cca(dune, dune.env) max differences in any method
are at magnitude 1e-15. This really does not matter, but since LC
scores are found first and WA scores are derived from those, LC
scores may be preferable. In old code, dbrda already used LC scores,
but cca, rda & capscale used WA scores.

R/centroids.cca.R

@@ -51,7 +51,7 @@
 {
     if (is.null(ord$CCA) || ord$CCA$rank < 1)
         return(NULL)
-    centroids <- centroids.cca(ord$CCA$wa, mframe, ord$rowsum)
+    centroids <- centroids.cca(ord$CCA$u, mframe, ord$rowsum)
     if (!is.null(ord$CCA$alias))
         centroids <- unique(centroids)
     ## See that there really are centroids