Estimate how many users a bad version reaches

A canary puts a new version on a small slice of traffic. Before yours starts, you want to know how far a bad version could spread. If the new version is bad, each task it handles fails visibly with a chance you can estimate from the task's type. The canary controller rolls the version back as soon as it has seen k visible failures, and no later task reaches the version.

Write expected_reach(tasks, k) that returns how many different users a bad version reaches on average.

Input and output:

• tasks: 1 to 3,000 tasks in arrival order, each [user, chance]. user is a non-empty string compared exactly. chance is a whole number from 0 to 100, the percent chance that this task fails visibly. Tasks fail independently of each other.
• k: a whole number from 1 to 200.
• Return a float. The tests accept any value within 0.000001 of the exact one.

Rules:

• Tasks reach the version in order until the rollback. The task that causes the k-th failure did reach the version, so its user counts.
• A user counts once, however many of their tasks reached the version. Each of those tasks can still fail.
• If fewer than k failures ever happen, every task reaches the version.