Unclear how to clean up from interrupted CreateCheckpoint

Related: #6015, #6233

### Expected behavior

CreateCheckpoint creates a clone db in a specified directory. You'd expect any of these to clean up from an interrupted CreateCheckpoint:
a) rm -rf the specified directory
b) DestroyDB the specified directory
And maybe
c) Do another CreateCheckpoint on the same directory

### Actual behavior

(a) and (b) don't work, because until completed, the db is in a tmp directory. (c) does work, but that's not intuitive or friendly because although CreateCheckpoint will clean up the tmp directory, it will fail if the specified db directory itself does exist. Good for atomicity, bad for clear deletion semantics.