Blender Git Loki

Cleanup: Use C++ types and lambdas for mesh normal calculation

- Use `threading::parallel_for` for multithreading, for a simpler API,
more readable and concise code.
- Use `Span` and `Array` (only internally, the public API is still C)
for safer, more automatic memory management.
- Since code is much less verbose, combine the callbacks into the
main function. Note that the accumulation code could be more concise
with `float3`, I just wanted to keep these changes minimal.

Differential Revision: https://developer.blender.org/D12402