[tinker] Propagate all uv run args from server startup command to backend workers

[SumanthRH]

What does this PR do?

Fixes #1234

After the fixes in this PR, I can successfully train with meta-llama/Llama-3.2-1B by passing an env-file to the server startup command :

 uv run --extra tinker --extra fsdp --env-file skyrl-train/.env.example -m skyrl.tinker.api  --base-mod
el meta-llama/Llama-3.2-1B   --backend fsdp 

(For context: See #1231 )

---

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no potential bugs to report.

View in Devin Review to see 4 additional findings.

[gemini-code-assist]

Code Review

This pull request refactors the logic for propagating uv run arguments from the API server to the backend workers, replacing the specific _get_parent_uv_accelerator_extras function with a more general _get_parent_uv_run_args and encapsulating command building in _build_uv_run_cmd_engine. While these changes improve modularity and robustness, a security risk has been identified: the implementation logs these uv run arguments and includes them in exception messages. As uv run arguments can contain sensitive information (e.g., secrets passed via --env), this could lead to sensitive data exposure in application logs. It is recommended to sanitize these logs to prevent such exposure.