chore(deps): upgrade runtime dependencies and CI workflow

[garrett4wade]

Description

Upgrade megatron-core, sglang, vllm (0.19.1), transformers, and related packages.
Pin deepep/deepgemm commits and lock trackio version for reproducibility.
Optimize the Dockerfile, add a uv_sync.sh helper, and remove integration tests
from per-PR CI (moved to nightly).

Type of Change

• ♻️ Refactoring

Key Changes

• Dependencies: Upgrade sglang, vllm (pinned 0.19.1), transformers, megatron-core
• vLLM compat: Remove unsupported args, fix worker extension for 0.19.1 API
• SGLang: Refactor launch_server/scheduler for latest sglang API changes
• Dockerfile: Slim image, optimize layer caching, fix uv install on Linux
• Archon: Fix Qwen3.5 precision args and model config
• Infra: Shrink port range, fix Ray utils, add engine blueprint guard
• CI: Remove integration tests from PR workflow (moved to nightly)
• Tests: Sync GRPO integration config, add agent_service controller test

Risk Areas

• vLLM worker extension rewrite — behavioral change in weight update flow
• SGLang scheduler refactor — changed server lifecycle management
• Dockerfile restructure — could affect image build caching

Test Commands Run

pre-commit run --all-files  # All hooks pass (skipped generate-cli-docs: no local Python)

Integration tests (SFT, GRPO) require GPU hardware — deferred to nightly CI.

Checklist

• I have read the Contributing Guide
• Pre-commit hooks pass (pre-commit run --all-files)
• Relevant tests pass; new tests added for new functionality
• Documentation updated (if applicable; built with ./docs/build_all.sh)
• Branch is up to date with main
• Self-reviewed via /review-pr command
• This PR was created by a coding agent via /create-pr
• This PR is a breaking change

Additional Context

43 files changed, 4461 insertions(+), 2911 deletions(-).
Lock file churn accounts for the majority of the diff (uv.lock, uv.vllm.lock).

[gemini-code-assist]

Code Review

This pull request introduces a new upgrade-deps skill and a comprehensive set of API compatibility checklists for core dependencies like megatron-core, transformers, sglang, and vllm. It updates several package versions in pyproject.toml and pyproject.vllm.toml, including a bump to megatron-core 0.17.0 and the addition of Python 3.12 version markers. Feedback identifies a missing architecture suffix in the Dockerfile base image and recommends consistent application of the Python 3.12 marker to the mbridge package across both project manifests.

[HT-Yuan]

@garrett4wade
Hi, I'm working on adding Muon optimizer support (PR #1270) which depends on megatron-core >= 0.17.

I noticed this PR hasn't been merged yet — is there any blocking issue?

[garrett4wade]

@HT-Yuan Hi, the upgrade of megatron-core should not break anything, but in this PR we'd like to upgrade other packages as well, which involves some changes of the integration code. I think #1270 is parallel to this PR though. We can find someone review this PR when you are ready.

[garrett4wade]

Baking in https://github.com/inclusionAI/AReaL/actions/runs/25416326579 . Merging.