Added nixl_ep installation option to Dockerfile

[dfarge]

What?

Added option to build nixl_ep (yet to be merged as of writing) by passing the flag --build-nixl-ep to the build-container shell file.

Why?

To ease use and development of the nixl_ep example.

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

[github-actions]

👋 Hi dfarge! Thank you for contributing to ai-dynamo/nixl.

Your PR reviewers will review your contribution then trigger the CI to test your changes.

🚀

[eranrs]

I have no comments on this. @dfarge, make sure to test it again before merging after #1043 is approved and merged.

[ovidiusm]

Dependency was merged: #1043

Please post this for review so that we can have it merged by code freeze (Friday December 12th). Thank you!

[michal-shalev]

/ok to test 3adcfeb

[michal-shalev]

/ok to test 947eea9

[michal-shalev]

/build

[dfarge]

I made some adjustments as per the pathing changes for where nixl_ep is now located.
I built a container with cuda 12.9 with the new pathing and am sanity checking it. I'll push when done.

Additionally, I see an issue when trying to build the dockerfile with nixl_ep and cuda 13.0, @michal-shalev is it a known issue?
What I get is gpunetio isn't being built with cuda 13.0, and that fails the nixl_ep installation.

[dfarge]

/ok to test 6be5ee8

[dfarge]

Added a commit to fix the gpunetio's cuda 13.0 error we were getting.
Did a sanity check on EOS and elastic.py runs well on 1 and 2 nodes.

[ovidiusm]

/build

[ovidiusm]

/ok to test 1b7909c