rustc: Accept PartialOrd/PartialOrdEq for Eq/Ord

[alexcrichton]

This is a transitionary step towards completing #12517. This change modifies the
compiler to accept Partial{Ord,Eq} as deriving modes which will currently expand
to implementations of PartialOrd and PartialEq (synonyms for Eq/Ord).

After a snapshot, all of deriving(Eq, Ord) will be removed, and after a snapshot
of that, TotalEq/TotalOrd will be renamed to Eq/Ord.

[pnkfelix]

I cannot tell from #12517: Is the plan to support deriving(PartialEq) and deriving(PartialOrd) when this is all done, (in addition to supporting deriving(Eq) and deriving(Ord), that is)?

If so, then this PR should be updated to include tests of deriving(PartialEq) / deriving(PartialOrd), right?

[alexcrichton]

That is the plan, yes. I have updated some examples in the gist of mine.

You won't be able to say deriving(Eq, PartialEq) on the same type, but you will be able to derive at most one of them (either one).

I could add some tests, but I figured that the common code path for Eq/PartialEq would cover it.

[pnkfelix]

Its not just about being confident that your code is correct; tests in scenarios like this can be useful as reference examples when someone other than the PR author is grepping the code base.

It also sounds like there is behavior regarding the interaction when erroneously deriving both that would be worth including a compile-fail test for.

[pnkfelix]

(but it also sounds like it would be premature to try to include the aforementioned compile-fail test now.)