Quick fix for issue #2958

Ensure when flatMap-ing, that we don't append duplicates.
Whether this is correct needs yet to be analyzed, but in general,
we don't enfore unique-ness of task results, and even when we evaluate
disjuct tasks, we can't assume that all results are disjuct, hence
using a strict Agg, which enforces uniqueness on append seems not
correct.

Here's my reasoning: If we have two tasks A and B, and B just forwards
the result of A as it's own result, both tasks return the same result.

main/eval/src/mill/eval/EvaluatorCore.scala

@@ -169,7 +169,11 @@ private[mill] trait EvaluatorCore extends GroupEvaluator {
 
     EvaluatorCore.Results(
       goals.indexed.map(results(_).map(_._1).result),
-      finishedOptsMap.map(_._2).flatMap(_.toSeq.flatMap(_.newEvaluated)),
+      Strict.Agg.from(
+        Loose.Agg.from(
+          finishedOptsMap.values.flatMap(_.toSeq.flatMap(_.newEvaluated))
+        )
+      ),
       transitive,
       getFailing(sortedGroups, results),
       results.map { case (k, v) => (k, v.map(_._1)) }