What Distributions are Robust to Indiscriminate Poisoning Attacks for Linear Learners?