Remove Sablono dependency, do templating internally

[tonsky]

Rum batches all updates and applies them all at once on requestAnimationFrame.

This doesn’t work well with DOM/React input fields and controlled values: they need re-render synchronously, otherwise you’ll see junk while typing in input field, selection will be broken, etc.

Sablono has special code that wraps every input in a wrapper that does that synchronous update directly in place, without waiting for next frame or anything. Problem is, this code adds complexity, is not entirely transparent, and doesn’t work very well. We’ve worked with r0man tried to fix it a few times (r0man/sablono#129, r0man/sablono#127, r0man/sablono#123) but ultimately we didn’t succeeded. In fact, I believe it is still broken somehow in 0.8.6, and that’s the reason why Rum stays on Sablono 0.8.1 instead.

Now, it all can be solved much easier if Sablono could call Rum re-render from input handlers. You wouldn’t need wrappers, I believe. Problem is, Sablono doesn’t know anything about Rum.

Solution I propose: remove Sablono dependency and implement templating entirely inside Rum. That would allow for much simpler and (hopefully) finally working inputs code.

Another benefit would be that server-rendering code would more closely match client-rendering code with regards to e.g. nil handling etc (r0man/sablono#132).

It will also allow us to fix some issues that are pending in Sablono as it seems not to be actively maintained, e.g. r0man/sablono#201