Experimental mesh light sampling improvements by caching the samples

[IntegratedQuantum]

While it does reduce block placement lag significantly (9 ms → 7.5 ms) to the point where it only takes 20% of the remaining latency on the render thread (most of the work is distributed). The main bottleneck (when placing white glowcrystals) is now light propagation with around ²⁄₃ of the latency.

However this comes at a cost of increased complexity and decreased generation throughput.

• It might make sense to switch between the old and new implementation based on the cache size to face count ratio to improve performance on LOD chunks.
• Cleanup the code, use enum instead of bool for the light parameter, pass it in from further up to reduce code duplication.
• Move the LOD switch to a separate PR, so we can check its impact independently of the cache changes.
• Make a throughput (chunk generation) and latency (placing blocks) comparison in various environments