Added baseline prompt and README example

[jamesbraza]

Used lmi/aviary to run a baselines on ether0-benchmark with gpt-4o.

I also realized we didn't open source our baselines prompt yet.

Here is the printed output:

In category 'functional-group' of 10 questions, average reward was 0.000.
In category 'molecule-completion' of 25 questions, average reward was 0.200.
In category 'molecule-formula' of 25 questions, average reward was 0.000.
In category 'molecule-name' of 25 questions, average reward was 0.080.
In category 'oracle-solubility' of 25 questions, average reward was 0.040.
In category 'property-cat-eve' of 25 questions, average reward was 0.400.
In category 'property-cat-safety' of 25 questions, average reward was 0.360.
In category 'property-cat-smell' of 25 questions, average reward was 0.320.
In category 'property-regression-adme' of 25 questions, average reward was 0.480.
In category 'property-regression-ld50' of 25 questions, average reward was 0.360.
In category 'property-regression-pka' of 25 questions, average reward was 0.280.
In category 'reaction-prediction' of 25 questions, average reward was 0.160.
In category 'retro-synthesis' of 25 questions, average reward was 0.000.
In category 'simple-formula' of 15 questions, average reward was 0.000.
Cumulative average reward across 325 questions was 0.206.

[Copilot]

Pull Request Overview

This PR adds a baseline prompt and example usage documentation for the ether0 benchmark while also updating dependency specifications to support the new baselines feature.

• Added a new constant (LOOSE_XML_ANSWER_USER_PROMPT) in model_prompts.py for XML-formatted SMILES answers.
• Updated pyproject.toml with new baselines dependencies.
• Extended README.md with benchmark instructions and a complete code snippet to run the evaluation.

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated no comments.

File	Description
src/ether0/model_prompts.py	Added a constant for XML-based SMILES answer prompts.
pyproject.toml	Added new dependencies required for running baselines.
README.md	Added detailed benchmark instructions and example evaluation snippet.

Comments suppressed due to low confidence (2)

src/ether0/model_prompts.py:121

• [nitpick] Consider adding a comment above this constant to explain its purpose and usage context for future maintainers.

LOOSE_XML_ANSWER_USER_PROMPT = (

README.md:177

• Ensure that the use of 'await' at the top level is supported in your ipython environment or include guidance to wrap asynchronous calls inside an async function.

results = await asyncio.gather(