bump oversampling defaults

[Jackmin801]

Higher oversampling default should make the averange run more efficient as it schedules more concurrent requests

---

Note

Medium Risk
Changes default RL run behavior by enabling oversampling (oversampling_factor=1.5) when not explicitly configured, which can increase compute usage and alter training dynamics. Also updates the generated config template’s recommended max_async_level, affecting throughput expectations.

Overview
Makes oversampling_factor default to 1.5 in RLConfig, so runs oversample unless users explicitly disable/override it.

Updates the prime rl init TOML template comments to reflect the new oversampling default and a lower recommended max_async_level (2 vs 4).

^{Reviewed by Cursor Bugbot for commit a48e7ee. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit a48e7ee. Configure here.}