Fix cross-crate inlining of static methods

[brson]

r? @pcwalton

This fixes translation of static methods with inline attributes, sourced from non-local crates. The problem is that maybe_instantiate_inline tries to look up the self type according to the self_did of the encoded inline method, and for static methods that did is just some random number unassociated with any type (afaict).

[graydon]

"some random number"? that's somewhat worrying. can we trap that condition?

[brson]

Sorry, I didn't literally mean 'random', I just mean that all method AST nodes have a def_id for the 'self' parameter, even if they are static methods.

[graydon]

Huh. Ok. I guess I was asking whether that ought to be Option<def_id> or such, but it's not critical, just .. nice to have these things cause a moment of pause about missed cases: Option sometimes causes the programmer to pause to think "wait, when can this be None"?

But that's kinda janitorial and shouldn't block this. This looks fine for now.

[brson]

It seems like it should, yes. I guess I should have asked plainly in the pull request whether I should make the proper refactoring. Since you are uncomfortable with it too, I'll keep at it and try to fix the AST.