[feature-request] Add support for exllamaV2

[shwu-nyunai]

Feature type?
Algorithm request

Proposal
Exllama V2 is one among the prominent¹ llm inference libraries out there.
GitHub Repo - https://github.com/turboderp/exllamav2

nyuntam's text-gen compression suite should extend the support for it as an engine.
We'd primarily want to support the quantisation scheme in exllamaV2 along with the supported models ootb ( conversion instructions here ) namely, EXL2;

¹Faster and Lighter LLMs: A Survey on Current Challenges and Way Forward