This required using a fork of Embree, newer LLVM version, unreleased ISPC
version and sse2neon directly from Git. Hopefully over time all the required
changes end up in official releases. For now we deviate from other platforms.
Based on contributions by Apple and Stefan Werner.
Ref D9527, D8237, T78710
Since OpenImageDenoise requires a CPU with SSE 4.1 or newer,
let the node act as passthrough on unsupported CPUs and display
a message in the node itself.
Running multiple instances of OIDN simultaneously can use dozens
of GBs of memory. Since OIDN is multithreaded internally, we can run
only one instance at a time and should not lose much performance.
Fixing T69006