Also fix snapshot_util() to work even if default_attr_ids is not a
proper lua array.

https://github.com/mpeterv/luacheck/pull/81#issuecomment-261099606

> If you really want to use bleeding-edge version you should get the
> rockspec from master branch, not a fixed commit ...
> The correct way to install from a specific commit is cloning that
> commit and running "luarocks make" from project directory. The reason
> is that running "install" or "build" on an scm rockspec fetches
> sources from master but uses build description from the rockspec
> itself, which may be outdated.

Call update_topline() to adjust the current viewport.

Closes #5597

Remove the inner for-loop; just use the column of the first in-line
match.

Restore cursor position and window "view", if command is cancelled.
This is how 'incsearch' works.

- Use a standard scratch buffer instead of a new 'buftype', functions
  like curbufIsChanged() already have special handling for scratch bufs.
- Cleanup some stuff from the previous merge.
- Add support for :smagic, :snomagic. Closes #5578

Add tests for undotree().

Helped-by: Björn Linse <bjorn.linse@gmail.com>

When "curhead" points to a valid head, the value of "newhead" is
meaningless (and really should be set to null). The undo state for
a buffer is _logically_ the enum:

  enum UndoState {
    CurrentHead(head),
    NewHead(head),
    EmptyTree
  }

nvim _represents_ this as: whenever `curbuf->b_u_curhead` is nonnull it
should be used as the current head, and `curbuf->b_u_newhead` is
ignored. If the there is a current head, then this will be redoed on the
next redo, and its parent will be undone on next undo. Only if
`b_u_curhead` is NULL, `b_u_newhead` will be used as the head to undo
(and it is not possible to redo). Also both can be NULL, to indicate an
empty undotree. (To be fair, this only strictly true when calling undo.c
from the outside, in some places _within_ a function in undo.c both
values might be meaningful)

Apparently `undotree()` breaks this non-abstraction, this _cosmetic_
issue can easily be fixed by `ex_substitute` also saving and restoring
`b_u_newhead`, but is doesn't reflect any error really how
`u_undo_and_forget` manipulates the _actual_ state of the undo tree.

During the live preview, the buffer-local 'modified' flag
should not be changed.