Expose k and t parameters in MCMC relocation

[fwilliams]

These parameters control the falloff curve for noise added to each Gaussian during optimization as a function of scale and opacity (more opaque Gaussians are considered more certain and undergo smaller perturbation during optimization). This PR doesn't change the default values of these but exposes them so they can be tuned as hyperparameters.

[matthewdcong]

It looks fine for the most part, but should we rename the variables to something more descriptive other than t and k?

[fwilliams]

It looks fine for the most part, but should we rename the variables to something more descriptive other than t and k?

I am not sure what to name these variables. It would end up being something verbose but not necessarily helpful (like noise_sigmoid_falloff_rate and noise_sigmoid_falloff_point neither of which are that helpful imo). The current names match those in eq.8 from the paper, which means you can refer back to that to see what they are. I can make the docs reflect this more clearly.

I don't expect these variables to be something a regular user would interact with. It's more for us to be able to tune default settings without having to recompile the code.

[matthewdcong]

OK, if it's mostly for developer use then we can just keep it as is for now. The comments outline the usage and are probably sufficient for dev purposes.