A huge poll that misses

You will work out exactly where a poll with non-response points, measure its bias, and then run the poll many times to see that a large sample does not remove the bias.

The idea. A share $p$ of the population supports a plan. Supporters reply with chance $r_y$ and the others with chance $r_n$. Among the replies, the share of supporters is $\dfrac{p\,r_y}{p\,r_y + (1 - p)\,r_n}$. Here every share is a decimal between 0 and 1.

Your tasks.

1. Finish reply_share(p, r_yes, r_no). Return the share of supporters among the replies.
2. Finish bias(p, r_yes, r_no). Return the reply share minus the true share.
3. Finish run_poll(asked, p, r_yes, r_no). For each person asked, decide at random whether the person supports the plan and whether the person replies. Return the share of supporters among the replies.