Correct typos

[ghabault]

• There is an error in the posterior_variance definition that should be divided by (1 - alpha_) otherwise the (1 - alpha) are cancelling each other out.
• torch.min, torch.max and torch.cumprod should be dim and not axis
• Detach the last timestep of the batch data in generate_data so that there is no need to do it after when calling generate_data