[Breaking Change v10] Nstep no longer returns "discounts"

[ymd-h]

We released cpprb v10 just now. (Binary will be uploaded to PyPI soon.)
ReplayBuffer with Nstep no longer returns "discounts" because users can always multiply with fixed gamma ** nstep.

For example, if we have a terminated trajectory , then 3-step targets become as follows;

As long as done are correctly calculated, sample["rew"] + (gamma ** nstep) * (1 - sample["done"]) * Q(sample["next_obs"]).max(axis=1) is fine.

If you have any questions, please feel free to ask me.

Ref: https://gitlab.com/ymd_h/cpprb/-/issues/137
Ref: https://ymd_h.gitlab.io/cpprb/features/nstep/