Comment 8 for bug 206406

The fix I have broke some other bits.
Basically, we still need to track down all the cases where we are holding a basis_tree longer than the containing working_tree's lock.

I thought there would be some easier/quicker workarounds, but it doesn't seem like it works.