Update infrastructure

[fasmat]

This fixes the build pipelines for gpu-post:

• Update zlib from v1.2.11 to v1.2.13
• Update build pipelines:
  • Use Ubuntu 22.04 instead of 20.04
  • Use Windows Server 2022 instead of 2016
  • Update Vulkan on Ubuntu from 1.2.154 to 1.3.231
  • Update CUDA on Ubuntu from 11.4 to 11.8 and fix installation script
  • Update Vulkan on macOS from 1.2.170.0 to 1.3.231.0 and fix installation script
  • Update CUDA on Windows from 11.2.2.46133 to 11.8.0.52206
  • Update Vulkan on windows from 1.2.154.1 to 1.2.182.0

[pigmej]

I checked again, and currently, Running tests fail with Vulkan SDK 1.3.231.1 on self-hosted mac m1.
It passes on 1.2.198.1

I'd, for now, downgrade to 1.2.x everywhere.

Besides that all good IMHO.

[pigmej]

The last remaining question would be why we have to use 1.2.198.1 on m1 while all others work with 1.3.x

But given that somehow m1 seems to be also super slow, (labels-count takes 12+minutes while other platforms <10m) then maybe the problem is "somewhere" else probably.

[pigmej]

These are few nit picks, we can do them here or in separate :)