You're right that we don't have a band solver (gbtrf or gbsv) now. We have had some other requests for this, so there's a possibility it will get implemented, but no firm plans. We can update you on this list if that changes.
Is your matrix is non-symmetric, so you need gbtrf instead of pbtrf (for a Hermitian positive definite matrix)? Do you ultimately need a solve, so doing gbsv instead of gbtrf & gbtrs is okay? Do you specifically need LU, or would some other factorization like QR also be okay? That might be easier than dealing with pivoting in a band matrix.