Allow setting the onevent callback per-call for m.redraw

[dead-claudia]

(Elevating this comment into an issue)

I could use the ability to set per-call the onevent callback for custom auto-redraw behavior on nodes rendered in that call only.

Also, that option could be used by redrawService to avoid needing the setEventCallback API in the first place, making it a little easier to integrate the two.

Note: this option would change vnode.events objects accordingly if onevent was already set for it.

(Yes, this implies a slight memory increase for vnodes with event handlers, but I doubt perf implications would be significant at all, since the event objects are persisted across redraws anyways.)

[dead-claudia]

This is all I need to integrate my SelfSufficient helper (a userland implementation of subtree redrawing) into Mithril's autoredraw system (and nullify the need for one of my existing API methods).

---

Side note: I'm relying on a few internals in that to actually do it right:

• vnode.dom, vnode.domSize, vnode.state, and vnode.events are all copied.
• I also have to manage vnode.instance directly on explicit redraw, so Mithril diffs against the most recent tree.

Also, I had to reimplement the following (amounting to ~30% of the code):

• Mithril's throttler (practically duplicated, but the logic is a little less centralized in mine)
• Mithril's auto-redraw mechanism (with a few modifications prohibiting simple adaptation)
• Mithril's onbeforeupdate logic (partially)
• These two functions (combined into one since I only needed it to call hooks)
• The auto-redraw hook and callback (as a separate method I'd ideally like to get rid of as part of this issue)

[dead-claudia]

One other thing I'd do as part of this is move the e.redraw checks to m.render itself instead of doing it in m.mount. This simplifies both m.mount and user onevent implementations and it saves a few bytes in the process.

[dead-claudia]

And one last thing: I would add a final render argument to oninit* and constructors so they don't have to depend on the global redraw instance.

* Only if we decide to keep stateful object components. (This was discussed on Gitter and referenced in #2293.)