feat: gp noise in discovery

[jeremy-baier]

Model Building

• build discovery likelihood with different settings
• includes the time domain SWGP using a custom segmentation ( from mercedes )
• this requires an addition to discovery, coordinating with pat
• set up dmgp, chromatic gp, ( dmx too )

Sampler set up

• set up a basic numpyro nuts sampler with tunable nuts settings
• pull in Dave + Pat’s code for optimization
• we probably need check pointing for the NUTS, which i can steal from the old outliers PR.

Add noise to model

• turn output of the NUTS into laforge core and/or even just a par.txt, chain.txt
• add noise to pint models

—————————————————————

Interfacing model building with notebooks

noise_run:
  model:
      include_rn:
          basis: interpolation/fourier
          Nfreqs: 100
          dt: 3
          psd: powerlaw
          kernel: sq_exp
      include_rn:
          basis: interpolation/fourier
          Nfreqs: 100
          dt: 3
          psd: powerlaw
          kernel: sq_exp
      inc_chromgp: false
  inference:
     likelihood: discovery
     sampler: NUTS
     n_iter: 2000
     n_warmup: 500
     n_checkpoint: 10