FEATURE REQUEST: Use GPUs for tasks that can utilise them.

[darcyabjones]

SignalP6 is very slow on CPU but reasonably quick on GPU.

Others like deepredeff, deepsig, deeploc, and signalp5 aren't prohibitively slow on CPU, but should be able to be sped up with a GPU.

This would probably require a separate install environment using GPU variants of deep-learning tools.
We'd also need to find a good way to make this customisable with configuration. I'm thinking that instead of a cpu_small kind of label, introduce a gpu_task label. In CPU environments set this to X-many cpus. In a GPU environment set this to use the GPU.

Could be a pain, but signalp6 is really slow without it, making it impractical to run for more than a few thousand proteins.

[darcyabjones]

Note, that Signalp6 can be sped up quite a lot simply by increasing the minibatch size. with the --bsize argument.
The default if 10 is super un-necessarily conservative. Especially for CPUs.
Bumping it up to 64 roughly triples speed.