Fix stale worker after refork: remove phase bump from mold_and_refork!

The phase bump on the mold candidate in mold_and_refork! had two
compounding bugs:

1. promote_mold filters by `w.phase == @phase`, so the bumped worker
   was excluded from selection — a different worker got promoted instead
2. After start_phased_restart increments @phase, the bumped worker's
   phase matched the new @phase, so check_workers skipped it — leaving
   one worker un-restarted with stale process state

Without the phase bump, promote_mold naturally selects the most
experienced worker (same one mold_and_refork! would have chosen), and
all workers are properly restarted during the phased refork.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: mlarraz

lib/puma/cluster.rb

@@ -322,9 +322,8 @@ def most_experienced_worker(workers = @workers)
       workers.max { |a, b| a.last_status[:requests_count].to_i <=> b.last_status[:requests_count].to_i }
     end
 
-    def mold_and_refork!(mold_candidate = most_experienced_worker)
+    def mold_and_refork!
       @mold&.term
-      mold_candidate.phase = @phase + 1 # cluster phase will catch up next loop; we want this one to be picked as a mold
       phased_restart(:refork)
     end
 
@@ -370,7 +369,7 @@ def setup_signals
           # wait before kicking off yet another phased restart
 
           current_interval_index += 1
-          mold_and_refork!(w)
+          mold_and_refork!
         end
       end