Reusing extract_answer_loose in accuracy_reward

[jamesbraza]

#6 forgot to use test arg, and when adding it just now, I realized:

1. accuracy_reward should reuse extract_answer_loose for test
   • This was due to the organic nature of our internal code base as many scripts, arising without a DRY'ing out
2. The baselines demo could use accuracy_reward directly

[Copilot]

Pull Request Overview

This PR refactors how the test flag is handled in reward extraction by reusing extract_answer_loose in the accuracy_reward function and updates the README to demonstrate its usage.

• Replaces duplicated logic with a concise ternary expression in rewards.py.
• Updates the README to leverage the new accuracy_reward function for baseline demos.

Reviewed Changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
src/ether0/rewards.py	Refactors answer extraction to use a ternary expression based on test.
README.md	Removes old extraction logic and shows usage of accuracy_reward.

Comments suppressed due to low confidence (2)

src/ether0/rewards.py:706

• [nitpick] Consider adding parentheses around the ternary expression to improve readability and clarify the execution order.

extract_answer_loose(content) if test else extract_thought_answer_strict(content, reasoning=reasoning)[1]

README.md:191

• [nitpick] Consider adding a brief comment explaining the use of strict=True in zip to clarify intent for future maintainers.

for prob_type, reward in zip(test_ds["problem_type"], rewards, strict=True):