CI: Manually generate pyflamegpu wheel for a specific python/cuda version

[ptheywood]

Occasiaonally producing "nightly" wheels has been useful between releases.

We can generate the full set of wheels by manually invoking the draft release CI workflow, but this will generate ~48 wheels, which are then only accessible for a short amount of time (24 hours?)

We could potentially make it possible FLAME GPU2 org members to manually run workflows to generate a wheel for a specific OS, CUDA and python version, using workflow dispatch inputs.

https://docs.github.com/en/actions/using-workflows/workflow-syntax-for-github-actions#onworkflow_dispatchinputs

This would save a lot of CI time, if these builds are more commonly required.