0.6.6

• Vastly improved garbage collection scheduling: previously Motoko runtime would do GC
  after every update message. We now schedule a GC when

  1. Heap grows more than 50% and 10 MiB since the last GC, or
  2. Heap size is more than 3 GiB

  (1) is to make sure we don't do GC on tiny heaps or after only small amounts of allocation.
  (2) is to make sure that on large heaps we will have enough allocation space during the next message.

  This scheduling reduces cycles substantially, but may moderately increase memory usage.

  New flag --force-gc restores the old behavior.

• Fix bug in compacting gc causing unnecessary memory growth (#2673)

• Trap on attempt to upgrade when canister not stopped and there are outstanding callbacks.
  (This failure mode can be avoided by stopping the canister before upgrade.)

• Fix issue #2640 (leaked ClosureTable entry when awaiting futures fails).