fix(cutting): fall back to a full order search when learning stops

Review found a plate the previous search planned that the tour now
refused. Left leads in on its left and right on its right, so right
straight after left crosses left and left straight after right crosses
right. Learning "right before left" then contradicted "left before
right" and the search returned ConstraintConflict, although cutting a
third part above them in between is safe.

A blocked approach only proves that one part cannot follow the parts
cut so far from that position, not a global order, so learned rules
stay a heuristic. Once nothing new can be learned, the remaining budget
now goes to a full search over every ready part, nearest first (the
search used before the tour). A rule contradicting an order already
required is still skipped rather than ending learning early.

docs/cutting-planner.md

@@ -88,7 +88,11 @@ before those", backs up to just before the earliest of them and re-plans the res
 from the tool position there; parts cut before that point are kept. An attempt
 stops backtracking after a stall of 8 x entries x contours expansions without
 getting further, so it learns instead of retrying every entry combination of the
-parts before it. When nothing new can be learned the result is a refusal.
+parts before it. A rule that would contradict an order already required is skipped.
+"Cut before" rules are a heuristic (a part blocked straight after another may be
+reachable via a third), so once nothing new can be learned the remaining budget
+goes to a full search that tries every ready part, nearest first; only when that
+also fails is the result a refusal.
 
 Candidates use native closest points, vertices, midpoints and circle angles in
 stable order, capped by `maxEntries`. Circle rounding, clamping, corner resolution