Draw-cache: Optimise empty cache-queued into single texture upload

[alexheretic]

This PR adds an optimisation targeting whenever the cache is empty & multiple new glyphs have been drawn in a call to cache_queued, e.g. the initial population of the cache & when it must be re-orded to fit all glyphs after a change.

In this case total in-use region of the texture will be saved in memory during the cache_queued(self, fonts, uploader) call and provided with a single callback to uploader. This minimises uploads to the gpu in common cases which should be faster.

cache_queued calls to a non-empty cache are not effected.

Benchmarks

The draw-cache v2 benchmarks simulate gpu uploads by busy waiting for 4us, this allows us to benchmark this improvement in a artificial way.

Non-populating benchmarks were not effected, as expected. Populating & reordering benchmarks were:

• multi_font_population_v2 -63.5%
• resizing_v2 -59.1%
• moving_text_thrashing_v2 -58.3%

Related to discussion in #120

[alexheretic]

@AlexCharlton I wonder how this change will perform in your profiled 100-glyph population use case, an improvement I'd hope?

[AlexCharlton]

I should have some time to do some more profiling this weekend, but I'm positive this change will affect that first-load time test positively :)