There is a potential bug after turning off incremental backup via incremental=(force_stop=true). Currently it only cleans up the resources in the running connection. So an attempt to use incremental after that call will not succeed. If one does a checkpoint and then attempts to use incremental, it will not succeed. If one closes and then reopens the WT connection the call will not succeed. However, if a crash happens before a checkpoint and then one reopens the WT connection, the call will succeed.
The reason is that the turtle file is not checkpointed currently as part of cleaning up the incremental resources.