Change Widget::update to mutate the Widget::State rather than return a new one. Move input capturing methods from Widget trait to UiCell.

[mitchmindtree]

This change addresses the problem highlighted in #554 where large widgets would be required to allocate an entire new Widget::State even for the slightest of changes. This is because the only way of updating the state was to return an entirely new state from the Widget::update method.

This PR changes the Widget::update method so that it no longer returns an Option<Widget::State> but instead provides a way of mutating the state via the State wrapper which is given via the UpdateArgs. All fields not relating to the unique widget state have been moved out of the State type and into a CommonState type. This allows us to use the State type purely as a "cell"-like wrapper around the WidgetState.

State::view provides immutable access to the Widget::State.
State::update allows for mutation of the Widget::State. When called, it sets a has_updated flag, allowing us to check whether or not the Widget::State was updated at all following the call of Widget::update and in turn allowing us to know whether or not we need to produce a new Element for the Widget.

This change also required moving the input capturing methods from the Widget trait to the UiCell, to be called within the Widget::update if necessary.

[mitchmindtree]

This really shouldn't be required, however I couldn't find any other way to appease the borrow checker. Perhaps there is something I'm missing? Would be nice if someone could review and see if they can get it working without the unsafe. @bvssvni ?

[bvssvni]

Checking now.

[bvssvni]

I've reduced it http://is.gd/Xu6jc0

[bvssvni]

Here is the solution: http://is.gd/LEoLyJ

You need to add another lifetime UpdateArgs<'a, 'b: 'a, W, C: 'a> and pass it in to pub state: &'a mut State<'b, W::State>. Notice that the second lifetime 'b inherits the first 'a such that 'b always outlive 'a.

This error happened because UpdateArgs moved, and the compiler only has 'a to reason about.

[mitchmindtree]

Woah, great stuff @bvssvni ! I tried multiple lifetimes but didn't consider bounding the second lifetime by the first, though it does make a lot of sense :) thanks! I'll implement the change now

[mitchmindtree]

All passing unsafe-less 😸 going to merge!

Published as v0.21.0.