Thanks for the question. Here are some thoughts.

- Linear Regression: I would use with low dimensional data (< 5) where relationships can be assumed to be linear and I only want to compute ATE.

- DML: I'd use when data are high dimensional (>10), functional relationships are non-linear, and I only want to compute ATE.

- Metalearners: Any case where functional relationships are non-linear and I don't just want ATE but also CATE and ITE.

Hope that helps! This is more art than science IMO. I'd be interested in hearing other thoughts on this.