RFC: allow special case Infinity grouping interval

[jacekkopecky]

This is a tentative PR for a feature that I think will benefit the vim-mode package (cf. atom/vim-mode#849). A change with groupingInterval of Infinity will group with any change coming after it (provided that one has non-0 groupingInterval), no matter how long it takes.

The rationale: in vim-mode, the command ciwabc<escape> will change (c) the current word (iw) to the text abc. An experienced VIM user will expect a single undo both to remove the abc and to restore the current word as it was, but it's also common that there's a break between the change command ciw and the typed text abc because they are cognitively different chunks of typing.

With this PR, vim-mode could group the deletion of the current word with the abc (provided that was typed fast enough for the default 300ms grouping interval), and experienced VIM users would get the benefit of undoing the semantically related changes together (makes sense in vim-mode), and still using the normal timed grouping of undo for text changes (makes sense in modern text editors).

To make sure those Infinity-grouped commands don't accidentally get grouped with changes that don't belong in a group, vim-mode can do an empty transaction with 0 grouping interval (as done in atom/vim-mode#849):

@editor.transact 0, -> # nothing

We can't easily use groupChangesSinceCheckpoint because there's no good way of knowing when to trigger it (at least I haven't found one).

I can add specs if you like the general idea.

[jacekkopecky]

That Travis error – illegal access – is that my fault?

[jamesgpearce]

I got the same on #94

[jamesgpearce]

PS I closed and re-opened min to kick travis to re-run (and it worked fine)