BUG? get the wrong value when logit_scale is 0

[shunshen93]

https://github.com/Dao-AILab/flash-attention/blob/v2.7.3/flash_attn/ops/triton/cross_entropy.py#L54

1. when logit_scale is 0.0, -inf * (0.0) is nan
2. when logit_scale is -1.0, -inf * (-1.0) is inf, inf - inf is nan
   so, when logit_scale is not bigger than 0, the result will be nan

and there is another complie error when n_cols is 1