Results 2 issues of ipiszy-x

**Describe the bug** ptr_LSE is never used in the fwd kernel.

bug
? - Needs Triage

**Describe the bug** The Blackwell FMHA fwd kernel causal mask implementation doesn't behave correctly when qlen != klen. **Steps/Code to reproduce bug** Unfortunately the FMHA verifier is also buggy because...

bug
? - Needs Triage