build: update build.py to pass versions as input parameter and convert version map to dictionary

[nvda-mesharma]

What does the PR do?

1. Replace TRITON_VERSION_MAP with a dictionary of key and values instead of remembering indexes throughout the script.
2. Add ability to pass backend version (start with vllm version) instead of hard-coding in the build.py
   example - https://github.com/triton-inference-server/vllm_backend/blob/mesharma-ci/ci/build/build_source.sh#L60

Checklist

• [ x ] PR title reflects the change and is of format <commit_type>: <Title>
• [ x ] Changes are described in the pull request.
• [ x ] Related issues are referenced.
• [ x ] Populated github labels field
• [ x ] Added test plan and verified test passes.
• [ x ] Verified that the PR passes existing CI.
• [ x] Verified copyright is correct on all changed files.
• [ x] Added succinct git squash message before merging ref.
• [ x] All template sections are filled out.
• Optional: Additional screenshots for behavior/output changes with before/after.

Commit Type:

Check the conventional commit type
box here and add the label to the github PR.

• [ x ] build
• ci
• docs
• feat
• fix
• perf
• refactor
• revert
• style
• test

Related PRs:

[Not ready for review, not needed to merge this change] triton-inference-server/vllm_backend#49

Where should the reviewer start?

all changes are contained within build.py script

Test plan:

1. Verified that the changes work with vllm_backedn ci build.
   Build: Trigger CI for new vllm_backend Triton releases vllm_backend#49
   All actions are successful and pipeline checks have also passed.
2. This is a no-op on release build process

CI Pipeline ID:

17157949

Caveats:

N/A

Background

I was unable to set any specific vllm version while triggering a build and test CI pipeline from vllm_backend repository due to the hard-coding in this build.py script.

Related Issues: (use one of the action keywords Closes / Fixes / Resolves / Relates to)

N/A

[oandreeva-nv]

Is it possible to start a new pipeline for this PR to check if it is goes well? + should we test newly added flags?

[nvda-mesharma]

I've tested the changes via this pipeline 18898001
All checks have passed here: triton-inference-server/vllm_backend#49
Pipeline link can also be found here

[oandreeva-nv]

I'm fine with these changes, just clarify help messages, as I mentioned here

If @mc-nv and @nv-kmcgill53 have no objections and there are no disturbances with our current CI, i.e. not the vllm specific, we should be fine.

[nv-kmcgill53]

We should be using the FLAGS for all of the versioning instead of the DEFAULT_TRITON_VERSION_MAP since each of the values in the map can be overwritten as passed arguments to build.py. You are defaulting the FLAG values to use the map as well. I don't really want bidirectional assignment from the FLAGS and the map, too confusing.

[nvda-mesharma]

Since there were some recent changes, I went ahead and ran a new pipeline to ensure that things were still working as expected
.../pipelines/20628717