patch 9.1.0116: win_split_ins may not check available room

patch 9.1.0116: win_split_ins may not check available room

Problem:  win_split_ins has no check for E36 when moving an existing
          window
Solution: check for room and fix the issues in f_win_splitmove()
          (Sean Dewar)

win_split_ins has no check for E36 when moving an existing window,
allowing for layouts with many overlapping zero-sized windows to be
created (which may also cause drawing issues with tablines and such).
f_win_splitmove also has some bugs.

So check for room and fix the issues in f_win_splitmove. Handle failure
in the two relevant win_split_ins callers by restoring the original
layout, and factor the common logic into win_splitmove.

Don't check for room when opening an autocommand window, as it's a
temporary window that's rarely interacted with or drawn anyhow, and is
rather important for some autocommands.

Issues fixed in f_win_splitmove:
- Error if splitting is disallowed.
- Fix heap-use-after-frees if autocommands fired from switching to "targetwin"
  close "wp" or "oldwin".
- Fix splitting the wrong window if autocommands fired from switching to
  "targetwin" switch to a different window.
- Ensure -1 is returned for all errors.

Also handle allocation failure a bit earlier in make_snapshot (callers,
except win_splitmove, don't really care if a snapshot can't be made, so
just ignore the return value).

Note: Test_smoothscroll_in_zero_width_window failed after these changes with
E36, as it was using the previous behaviour to create a zero-width window.
I've fixed the test such that it fails with UBSAN as expected when v9.0.1367 is
reverted (and simplified it too).

related: #14042

Signed-off-by: Sean Dewar <6256228+seandewar@users.noreply.github.com>
Signed-off-by: Christian Brabandt <cb@256bit.org>