Bugfix/3943 docker container improvements

[Callum17]

Checklist:

• [✓] I have read the Contributing guidelines.

• less verbose docker compose

• drops containerized virtualenv in favor of non root runtime user

• slightly smaller runtime image

• moved all dependencies to builder pattern - so it should be easier to build and test newer dependencies

Refactored this PR last moment as the original exllamav2 3943 issue was fixed here:
#4095

We should be able to close this issue:
#3943

My only question might be around the bump from flask_cloudflared==0.0.12 to flask_cloudflared==0.0.14 in the openai extension. Had to be bumped to prevent conflicts with the api extension. I still need to test this change. Functionally should be no different from the version installed by the current docker compose setup - unlike the current setup this builder pattern will explicitly fail if any constraints are violated in any of the requirements files.

Also worth noting that a number of conflicts were found against the (outdated?) requirements.txt in oobabooga/GPTQ-for-LLaMa. Conclusion was that it can be ignored.

inference successfully tested with:
ctransformers
exllama
exllamav2
llama.cpp
AutoGPTQ

[LoopControl]

Suggested change

	# ignore custom container envrionment
	# ignore custom container environment

[oobabooga]

Could you merge the dev branch and check if the updated Dockerfile still works?

A relevant detail is that the OpenAI API is now the default, and it listens on port 5000 only (5005 is not used anymore).

I don't know how to use Docker properly, but if you find that your updates still work, I can merge the PR based on your feedback.

[Callum17]

I've got an updated copy working against current HEAD of main branch. Still needs a minor tidy and the dev merge, but it's pretty much there.

Will look to test and update this in the next day of two when I get time.

[oobabooga]

Awesome @Callum17, thank you.

[Callum17]

Think GitHub auto closed this when I was aligning with the dev branch.

Opened a clean PR here:
#4768

There are some issues (not created by me!) - most notably with GPTQ-for-Llama conflicting with the core package dependencies. I suspect we will start seeing issues in non docker installs if these aren't resolved.