Try to fix spurious circular task errors (#6062)

Fixes #6060

My guess is that the problem is the `LinkedHashSet` in
`Cacher.taskEvaluationStack` is actually a mutable collection shared
between different threads. That could potentially cause all sorts of
problems, so this replaces it with a immutable `VectorMap` to preserve
ordering and provide lookups in an immutable fashion

---------

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

Author: lihaoyi

core/api/src/mill/api/internal/Cacher.scala

@@ -1,6 +1,6 @@
 package mill.api.internal
 
-import collection.mutable.LinkedHashSet
+import scala.collection.immutable.VectorMap
 import scala.collection.mutable
 import scala.util.DynamicVariable
 import scala.quoted.*
@@ -12,7 +12,8 @@ trait Cacher extends mill.moduledefs.Cacher {
     if (Cacher.taskEvaluationStack.value.contains((c, this))) {
       sys.error(
         "Circular task dependency detected:\n" +
-          (Cacher.taskEvaluationStack.value.toList ++ Seq((c, this)))
+
+          (Cacher.taskEvaluationStack.value.keys ++ Seq((c, this)))
             .map { case (c, o) =>
               val taskName = c.value.split("\\.|#| ").filter(!_.startsWith("$anon")).last
               o.toString match {
@@ -24,19 +25,16 @@ trait Cacher extends mill.moduledefs.Cacher {
       )
     }
 
-    try {
-      Cacher.taskEvaluationStack.value.add((c, this))
+    Cacher.taskEvaluationStack.withValue(Cacher.taskEvaluationStack.value ++ Seq((c, this) -> ())) {
       cacherLazyMap.getOrElseUpdate(c, t).asInstanceOf[T]
-    } finally {
-      Cacher.taskEvaluationStack.value.remove((c, this))
     }
   }
 }
 
 private[mill] object Cacher {
-  // Use a LinkedHashSet for fast contains checking while preserving insertion order
+  // Use a VectorMap for fast contains checking while preserving insertion order
   private[mill] val taskEvaluationStack =
-    DynamicVariable[LinkedHashSet[(sourcecode.Enclosing, Any)]](LinkedHashSet())
+    DynamicVariable[VectorMap[(sourcecode.Enclosing, Any), Unit]](VectorMap())
   private[mill] def withMacroOwner[T](using Quotes)(op: quotes.reflect.Symbol => T): T = {
     import quotes.reflect.*
     // In Scala 3, the top level splice of a macro is owned by a symbol called "macro" with the macro flag set,