DEScrypt-opencl lacks a done() function, thus leaks memory

[magnumripper]

https://github.com/openwall/john/pull/4354#issuecomment-698077601

This is not much of a problem in real use, leak seen with --stress-test --format=DEScrypt-opencl.

I will personally never look into this very format, the code isn't compatible with my brains.

[solardiz]

This isn't detected during all-formats --test (including CPU and OpenCL) by gcc's LeakSanitizer on Debian 13 as tested in #5893. The format does still lack done, but does it need it? Maybe --stress-test is somehow special?

[magnumripper]

The format does still lack done, but does it need it?

If it has its own autotune, I assume it deallocates between tries, at least. But anyway, before we started deallocating in done(), we often ran out of GPU memory during all-format tests. So I think we should fix it. And --stress-test is also a reason.