Optimize style value resolution with handle-based approach

[NickGerleman]

Summary:
This PR is an optimization attempt during work on the calc() feature: facebook/yoga#1874.
It aims to reduce the regression effect on layout operations after extending the size of StyleLength and StyleSizeLength classes.

Changelog:

This PR introduces a private resolve(StyleValueHandle handle, float referenceLength) function to Style.h that resolves its property values directly from StyleValueHandle and skips creating intermediate objects.

I am aware that this duplicates the resolving functionality that already exists in StyleLength.h and StyleSizeLength.h, and may not be the cleanest solution. Optimization could go even further and totally drop the resolve() function from StyleLength.h and StyleSizeLength.h, and replace all usages with the newly introduced functions. But I think the performance boost is not high enough to justify that. Also it won't be the best API for Yoga library consumers - that's why I decided to keep them. However, I am opening this PR to start a discussion and would really like to hear opinions about it.

Benchmarks:

Results were averaged over 20 runs.

This does not bring much gain in terms of performance when compared to the main branch.

However, it introduces a significant boost for the calc() and YGValueDynamic work.

Results before these optimizations:

And after:

cc NickGerleman

X-link: facebook/yoga#1922

Reviewed By: zeyap

Differential Revision: D99006553

Pulled By: NickGerleman

[meta-codesync]

@NickGerleman has exported this pull request. If you are a Meta employee, you can view the originating Diff in D99006553.