Add target that samples from Stan with full CLI

[WardBrian]

This was brought up by @bgoodri in the Stan meeting a couple weeks ago, and should hopefully be helpful for our own testing.

It adds a new target, examples/stan_cli that can run BridgeStan models. Unlike the pre-existing example that uses bridgestan, this only runs adaptive WALNUTS, and lets you customize all the current tuning parameters. I added CLI11 for this target to facilitate that.

Here's the help output

$ ./examples/stan_cli --help

Run WALNUTs on a Stan model 


./examples/stan_cli [OPTIONS] model [data]


POSITIONALS:
  model TEXT:FILE REQUIRED    Path to the Stan model library 
  data TEXT:FILE              Path to the Stan model data (optional) 

OPTIONS:
  -h,     --help              Print this help message and exit 
          --seed UINT [1823896191]  
                              Random seed 
          --warmup INT [1000]  
                              Number of warmup iterations 
          --samples INT [1000]  
                              Number of samples to draw 
          --max-depth INT [10]  
                              Maximum depth for NUTS trajectory doublings 
          --max-step-depth INT [8]  
                              Maximum depth for the step size adaptation 
          --max-error FLOAT [0.5]  
                              Maximum error allowed in joint densities 
          --init FLOAT [2]    Range [-init,init] for the parameters initial values 
          --mass-init-count FLOAT [1.1]  
                              Initial count for the mass matrix adaptation 
          --mass-iteration-offset FLOAT [1.1]  
                              Offset for the mass matrix adaptation iterations 
          --mass-additive-smoothing FLOAT [0.1]  
                              Additive smoothing for the mass matrix adaptation 
          --step-size-init FLOAT [0.5]  
                              Initial step size for the step size adaptation 
          --step-accept-rate-target FLOAT [0.666667]  
                              Target acceptance rate for the step size adaptation 
          --step-iteration-offset FLOAT [2]  
                              Offset for the step size adaptation iterations 
          --step-learning-rate FLOAT [0.95]  
                              Learning rate for the step size adaptation 
          --step-decay-rate FLOAT [0.05]  
                              Decay rate for the step size adaptation 
          --output TEXT:PATH(non-existing) 
                              Output file for the draws 

[bob-carpenter]

Now that you tracked down where the NaN issue was coming from, this looks good to go. I didn't extensively test all the input options were getting passed correctly.