D5415: rust: changed Graph.parents to return [Revision; 2]

[Yuya Nishihara]

> Another point is that there is no firm guarantee that a valid revision occurs before `NULL_REVISION`. I've found examples in `revlog.c` where the second parent is (seemlingly on purpose) not ignored if the first is `NULL_REVISION`, but I wouldn't want at this point to impose it from the `parents()` method right away.

Curious. Can you point out which? I think many things would be broken if
p1 is null but p2 isn't.