Document how to build ctranslate2 with zenddn

[commandline-be]

🚀 The feature, motivation and pitch

Having an on-site translation engine is both practical and privacy friendly to that end I sought to have a GPU accelerated translator.

Attempting to run argostanslate which builds on ctranslate2 I ended up with what seems like a dead end for an AMD GPU (Radeon VII) Digging a bit deeper I found agrostranslate > opennmt > ctranslate2 (supports CUDA 100% not AMD 100%) < onednn (supports AMD)

For reference

OpenNMT/CTranslate2#1072
ROCm/ROCm#2534

I'd appreciate your informed feedback.

Alternatives

alternatives considered and tried
use rocm 6.x

Additional context

Because I lack the background to progress thru trial and error on this matter I'd like to understand what is possible with ZenDNN/ZenDNN-pytorch to recompile and maybe counter any compatibility challenges