Assume a reduction of complexity as dimensionality increases through prior

## Issue

For our current ad-hoc priors for the RF calibration parameters
https://github.com/CliMA/CalibrateEmulateSample.jl/blob/eb262cb2cad654ca46634e8fa3c158ace1e9a225/src/RandomFeature.jl#L246

It may make sense to draw some inspiration from work such as [this](https://arxiv.org/pdf/2402.02229) where one takes less rough priors as (effective) dimensionality increases. This may make parameter spaces more robust to changing problem dimension