[bug] Causal mask is skipped if no att_mask is passed

[blefaudeux]

🐛 Bug

In https://github.com/facebookresearch/xformers/blob/main/xformers/components/attention/scaled_dot_product.py#L78
if no att_mask is passed, then an hypothetical causal mask is completely skipped

[blefaudeux]

cc @SeanNaren, thanks for the heads up, fixing that asap

[blefaudeux]

cc @dianaml0, maybe the reason why you were having a trouble comparing dense and blocksparse ? This was pretty wrong