> In sample() the selected set can be initialized to {n}

I originally used the {n} approach but it was less clear and it lead to a re-selection rather than undoing the rounding.  That would change the output.

> I like Tim's suggestion about import-time patching.

Sorry, but there's a limit to how much I'm willing to garbage-up the code over this issue.

> In choice() I would write the condition as "i == n > 0" to 
> avoid indexing with negative index

I'll write that as "'i == n and n > 0" which reads better and runs faster in the common case (look at the disassembly of each).

Attaching a revised patch.

> here's another timing variation:
>
>    i = int(random() * n)
>    return seq[i - (i == n)]

This ran a little slower than the conditional approach.