fix(exploration): typo self.ago -> self.algo in sample_func_logits

[jmtoepperwien]

Summary

In MightyExplorationPolicy.sample_func_logits, the sac flag passed to sample_nondeterministic_logprobs used self.ago instead of self.algo.

# Before (broken)
log_prob = sample_nondeterministic_logprobs(
    z=out[1], mean=out[2], log_std=out[3], sac=self.ago == "sac"
)

# After (fixed)
log_prob = sample_nondeterministic_logprobs(
    z=out[1], mean=out[2], log_std=out[3], sac=self.algo == "sac"
)

Impact

Because self.ago does not exist, the expression always evaluated to False, so the tanh-squash log-probability correction was never applied for SAC in this code path — even though SAC requires it. This caused old log-probs stored in the rollout buffer to be plain Gaussian log-probs without the tanh correction, introducing a systematic bias in the PPO importance-sampling ratio.