rebase: when collapsing, p1 == dest, so use the former only `dest` is the destination we're rebasing onto, which is always the same as `p1` when using `--collapse`. This lets us simplify a bit. Differential Revision: https://phab.mercurial-scm.org/D9073

rebase: remove redundant isinmemory argument from _origrebase() The argument is redundant with `rbst.inmemory`, which makes it unclear what to do if they were somehow different. Differential Revision: https://phab.mercurial-scm.org/D9070

largefiles: prevent in-memory merge instead of switching to on-disk I enabled in-memory merge by default while testing some changes. I spent quite some time troubleshooting why largefiles was still creating an on-disk mergestate. Then I found out that it ignores the callers `wc` argument to `mergemod._update()` and always uses on-disk merge. This patch changes that so we raise an error if largefiles is used with in-memory merge. That way we'll notice if in-memory merge is used with largefiles instead of silently replacing ignoring the `overlayworkingctx` instance and updating the working copy instead. I felt a little bad that this would break things more for users with both largefiles and in-memory rebase enabled. So I also added a higher-level override to make sure that largefiles disables in-memory rebase. It turns out that that fixes `run-tests.py -k largefiles --extra-config-opt rebase.experimental.inmemory=1`. Differential Revision: https://phab.mercurial-scm.org/D9069

merge: add a back_out() function to encapsulate update() I've sent several earlier patches adding `merge.clean_update()`, `merge.merge()` etc, one function for each use case. This patch continues that work. I plan to hide the complex `update()` eventually. Differential Revision: https://phab.mercurial-scm.org/D9064