D3896: copies: handle case when both merge cset are not descendant of merge base

[Yuya Nishihara]

Can you add some tests that make both `dirtyc1` and `dirtyc2` set, and
trigger copy tracing?

> Since we can have both dirtyc1, and dirtyc2 true, I am not sure whether
> this else should be turned into it's own if statement?

No idea. I doubt if the current copy tracing can handle such cases. From
my vague memory, the current algorithm relies on the fact that the pseudo
`base` revision exists in between the `c1` and the `c2`.