Performance: Cache results of hb_shape calls

[bryphe]

As can be seen from the profile here:

We spend a lot of unnecessary time shaping text over and over. We should be caching or memoizing this call - the shape results are constant for a (font family, text) tuple.

[bryphe]

The expensive call is here:
https://github.com/bryphe/revery/blob/fb0bb59faaa325245c0a669a4ecb06af5c6b9292/src/UI/FontRenderer.re#L43

Work needed:

• We should wrap Fontkit.fk_shape in a wrapper, like FontRenderer.shape(font, text).
• We should memoize the result of this, like we do for the getTexture call: https://github.com/bryphe/revery/blob/fb0bb59faaa325245c0a669a4ecb06af5c6b9292/src/UI/FontRenderer.re#L34
• We should replace the calls to Fontkit.fk_shape in both FontRenderer.re and TextNode.re with our cached / memoized version.