Consolidate simple_trainer.py and simple_trainer_mcmc.py

[brentyi]

Not fully tested yet, but here's a config structure suggestion to make this API possible: python simple_trainer.py {default,mcmc}.

python simple_trainer.py --help:

Default densification: python simple_trainer.py default --help

MCMC densification: python simple_trainer.py mcmc --help

[brentyi]

Pushing us in the direction of nerfstudio (splatfacto vs splatfacto-big, etc), this also makes it easier to have different sets of standard hyperparameters.

You can just add a new key/value pair to this dictionary:

    # Config objects we can choose between.
    # Each is a tuple of (CLI description, config object).
    configs = {
        "default": (
            "Gaussian splatting training using densification heuristics from the original paper.",
            Config(),
        ),
        "mcmc": (
            "Gaussian splatting training using densification  from the paper '3D Gaussian Splatting as Markov Chain Monte Carlo'.",
            Config(init_opa=0.5, init_scale=0.1, strategy=MCMCStrategy()),
        ),
    }

[liruilong940607]

LGTM! Thanks Brent!