vector layer: transpose / optimise the style buffer

[adam-ce]

try to transpose the style buffer, i.e, instead of using
style1zoom0, s1z1, s1z2.. s1z20, s2z0, s2z1..
we could try
s1z0, s2z0, s3z0, .., s1z1, s2z2, s3z2, ..

additionally, we could try to sort according to usage frequency such that often used styles are closer together. but this might be hard..

the goal here would be to maximise cache hits.

[adam-ce]

i just tested the potential:
without loading the style (no texture fetch, resulting in nothing being rendered): 61.3 ms
default (loading the style as it is now): 66.7 ms

so the upper bound for this change is ~9% with the current rendering algorithm (smoothstep based half-spaces). reevaluate later once other perf improvements are in place.