fix_caw: A fixed point iteration for CAW.
In dcgerard/vicar: Various Ideas for Confounder Adjustment in Regression

A fixed point iteration for CAW.

fix_caw(
  w_current,
  Z2hat,
  scale_val,
  rotate_out,
  weight_func = ash_wrap,
  weight_args = list(),
  min_scale = 0.8,
  return_val = c("params", "full")
)

`w_current`	A vector of weights. The probability given Z2hat that each coefficient is null. These should all be between 0 and 1.
`Z2hat`	A matrix. The current estimates of the unobserved confounders.
`scale_val`	A positive numeric. The variance inflation term.
`rotate_out`	The output from `rotate_model` with `degrees_freedom` added to the list.
`weight_func`	The function that calculates the weights given `w_current`. So far, this only supports functions that take as input three quantities: `betahat`, `sebetahat`, and `degrees_freedom`
`weight_args`	Additional arguments to pass to `weight_func`.
`min_scale`	A scalar. The minimum value that `scale_val` is allowed to take. In practice, you wouldn't think that shrinking the variance is a good idea. But a value of 1 seems to stringent. So I set this to 0.8 to give you some wiggle room. Most of the time `scale_val` is usually estimated greater than 1.
`return_val`	Should we return just the parameters of interest (`"params"`), or the full shebang (`"full"`)? Setting `return_val = "params"` is used mostly for the optimization and returns only `w_current`, `Z2hat`, and `scale_val`.