OpenCL_Brute bug: GPU Acceleration for Cardano seeds of length 15, 21 and Trezor-24 fails to find the seed (Only relevant to Cardano)

[3rdIteration]

There is actually an obscure bug in the upstream opencl_brute that the Cardano implementation uncovered.

Basically for DKlengths > 64, the second block (so the last 32 bytes if you are doing a dklen of 96 like in Cardano) will be incorrect if the salt length isn't an even multiple of the 8. (So likely an issue in padding) (Trezor has a bug with Cardano where the a Trezor T with a 24 word seed incorrectly includes the checksum in the entropy, leading to 33 bytes of entropy... causing this bug to manifest)