Add DAPO Recipe

[tyler-griggs]

This issue tracks progress towards implementing support for the DAPO recipe and launching a reproduction run.

• Support token level loss: Support token-level loss, make default #90
• Support clip higher
• Support overlong filtering [DAPO] Add support for overlong filtering #111
• Support dynamic sampling [trainer/algorithm] Implement DAPO and Polaris style dynamic sampling + add DAPO docs + example #130
  • Polaris-style
  • DAPO-style
• Customize compute_adv more flexibly with registry [Trainer] Support registering custom advantage estimators #115

[SumanthRH]

This issue is closed after #130