Fix eval_start_step logic with eval_interval

[zxhe-sean]

Description

This PR resolves the logic conflict and confusion when both eval_start_step and eval_interval are specified.

• Why is this change being made:
  Previously, when users set both eval_start_step and eval_interval, the evaluation logic had conflicting behaviors—either waiting until the end of the first full interval before starting or misaligning the initial evaluation step with subsequent intervals.
• The problem being solved and relevant context:
  Standardizing the behavior so that evaluation explicitly starts at eval_start_step and subsequently follows eval_interval steps ((step - eval_start_step) % eval_interval == 0).
  • Users have a clean, intuitive way to decide when evaluation first starts and how frequently it repeats.
  • If users want the legacy behavior (e.g., eval_interval=10 evaluating at steps 9, 19, 29...), they can achieve it under this unified model by setting eval_start_step=9.

FIXES: b/535641549

Tests

Tested locally

Checklist

Before submitting this PR, please make sure (put X in square brackets):

• I have performed a self-review of my code. For an optional AI review, add the gemini-review label.
• I have necessary comments in my code, particularly in hard-to-understand areas.
• I have run end-to-end tests tests and provided workload links above if applicable.
• I have made or will make corresponding changes to the doc if needed, including adding new documentation pages to the relevant Table of Contents (toctree directive) as explained in our documentation.

[gemini-code-assist]

Caution

The consumer version of Gemini Code Assist on GitHub has been sunset. All code review activity has officially ceased.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!