Explicit dims in model config

[wd60622]

The model config has implicit dim which reduces model customization

For instance, in model config of the current MMM has implicit dims for variables. For instance, the logistic saturation variables assume to be "channel" but are specified

model_config = {
    ...,
    # Currently written
    "saturation_lam": {"dist": "HalfNormal", "kwargs": {"sigma": 1}}, 
    "saturation_beta": {"dist": "HalfNormal", "kwargs": {"sigma": 1}}, 
    # But is actually
    # "saturation_lam": {"dist": "HalfNormal", "kwargs": {"sigma": 1}, "dims": "channel"},
    # "saturation_beta": {"dist": "HalfNormal", "kwargs": {"sigma": 1}, "dims": "channel"},
    ...,
}

However, this doesn't allow for common parameter between channels. For instance the following would suggest a common shape but different scale of curve:

model_config = {
    ...,
    "saturation_lam": {"dist": "HalfNormal", "kwargs": {"sigma": 1}}, 
    "saturation_beta": {"dist": "HalfNormal", "kwargs": {"sigma": 1}, "dims": "channel"}, 
    ...,
}

Related to #734

Nested / Hierarchical Parameters

In a larger model config, dimensions are required in order to specify differences

For instance,

model_config = {
    ..., 
    "beta": {
        "dist": "Normal", 
        "kwargs": {
            "mu": {"dist": "Normal", "kwargs": {"mu": 0, "sigma": 1}, "dims": "geo"},
            # Very different then this!
            # "mu": {"dist": "Normal", "kwargs": {"mu": 0, "sigma": 1}, "dims": "channel"},
            "sigma": 1,
        }, 
        "dims": ("channel", "geo"), 
    }, 
    ..., 
}

Related to #708