set_reference should copy the passed dataframe in most cases

[tms-bananaquit]

One thing we hadn't considered so far is that the user can pass X as the reference data set and then modify it. This behavior seems undesirable. The detectors ought to store a copy. It'll increase the memory footprint, but this doesn't seem avoidable.

I don't believe this is a problem for any(?) of the streaming detectors, because they shouldn't actually be storing the data directly, but that should also be checked and corrected.

• CUSUM is appending values internally, but as these are one-dimensional, it doesn't seem like a problem. It looks like list.append is creating a copy of the value and sticking it on the tail.
• MD3 needs updating.
• The other detectors aren't storing the values directly.