Cray libsci fails on large problems

**Describe the bug**
On a new Cray Intel sapphire rapids machine, qmcpack was compiled with gnu compilers + Cray libsci. So far as I can tell, this is like MKL, but Cray. The code compiles and passes 100% of deterministic tests. With batched drivers on a problem containing ~400 electrons, the code runs as expected and gives good results that agree with previous runs on a different machine using legacy drivers. However, on a much larger problem containing > 2800 electrons, VMCBatched segfaults immediately. A backtrace suggests the issue is related to the implementation of dgetrf, which is a LAPACK routine to compute the LU decomposition. This is handled by libsci.

I then recompiled the code using gnu + mkl, and, again, all deterministic tests pass 100%. And it actually works for small and large problems, too.

This isn't really a bug _per se_. But this caused a lot of consternation and time spent debugging. So I mostly wanted to document for posterity.