Update README.md about SkyRL-v0 reproduction

[caoshiyi]

Many people are trying to reproduce SkyRL-v0's results, temporarily trying to direct them to the old commit for exact reproduction.

[gemini-code-assist]

Code Review

This pull request updates the README.md file to include instructions for reproducing the SkyRL-v0 results. My feedback focuses on improving the clarity and user-friendliness of these instructions by suggesting better formatting and recommending the use of a Git tag for a more stable reference.

[gemini-code-assist]

For better clarity and to make it easier for users to reproduce the results, I suggest a few improvements:

• Improve formatting and wording: It's a good practice to wrap commands and commit hashes in backticks to format them as code. Providing the direct git command is also more user-friendly.
• Use a Git tag: For long-term reproducibility, consider creating and referencing a Git tag (e.g., skyrl-v0-repro) instead of a raw commit hash. Tags are more descriptive and stable references.

Here's a suggested change that incorporates the improved wording and formatting. Creating and using a tag would be an additional improvement.

Suggested change

	- [`skyrl-agent`](./skyrl-agent): Our agent layer for training long-horizon, real-world agents. Contains code for [SkyRL-v0](https://novasky-ai.notion.site/skyrl-v0). For exact reproduction of SkyRL-v0 results, please checkout to commit a0d50c482436af7fac8caffa4533616a78431d66. New code on SWE Agent training will be updated soon!
	- [`skyrl-agent`](./skyrl-agent): Our agent layer for training long-horizon, real-world agents. Contains code for [SkyRL-v0](https://novasky-ai.notion.site/skyrl-v0). To reproduce the SkyRL-v0 results, run `git checkout a0d50c482436af7fac8caffa4533616a78431d66`. New code on SWE Agent training will be updated soon!