Prerequisites

• I am using the latest version of Locust
• I am suggesting a new feature, not asking a question

Description

According to current implementation/documentation, if weights are given to tasks, tasks are stored in the list duplicated by the amount given as weight, and then the task is selected from the list by using Python's random.choice.
This, however, has two shortcomings:

• Inserting big weights can lead to unreasonably big datastructures
• Only integers can be weights

It should be fairly easy to rewrite the gen_tasks_from_base_classes function and how it's output is handled in TaskSet class to store the weights separately and pass them together with the list of tasks to random.choices in gen_next_task.

This will allow to store arbitrarily large weights without duplicating tasks in integer or float as well as potentially enable the functionality to modify task weights at runtime by modifying the list storing the weights.

I could try to implement this with a PR if this change is desired.