Ignoring the whole question of B-movie survival simulator vs. survival manual survival simulator (I don't see a "market" for a second survival manual survival simulator):
It is not possible to do a tracking fork of C:DDA. (This is what did in C:BrightNights). The conscious experience budget required to explicitly verify all commits is excessive (as a full-time job I could do maybe 10% of the daily commit rate).
Cherrypicking technical design decisions is eminently practical, and I expect that to go both ways. (I recognized some of the map breakage back pre-0.0.D as comparable to what I predicted my pre-0.2.0 changes to force correct save/load cycling would do to an existing save.) That said, new/reimplemented features are going to be arriving at a glacial rate even after the overmap and map representation issues no longer block a whole-Earth overmap system. [I don't see that actually happening, but the failure should be lack of CPU and hard drive space, not "can't even represent it"].
I don't expect to take radical risks with gameplay changes before the Socrates' Daimon assistant program isn't vaporware. I'll re-label the 0.2.0 download as Przybylski's Star after the 0.2.1 release, due to the coverage issue letting the modern MSVC++ handling of printf specifiers reach there.
One of my priorities is that any "stable" release (and indeed, almost all unstable checkpoints) should be valid to fork from.