Resolve virtual base functions (super.foo()) correctly

[kinke]

Don't perform a vtable lookup for those, resolve them directly instead.
Fixes issue #1450, where calling the base method in the overriding method resulted in an infinite loop and corresponding stack overflow.

[kinke]

We have the same check for DelegateExp, the only other place where we call DtoVirtualFunctionPointer().

Test will follow once merged. ;)

[kinke]

Updated as I figured the actual reason, not just the symptom. This was just begging for more issues. See my monologue in #1450. ;)

[kinke]

I'm planning to extend the direct-sret-construction to return statements as well, not just variable declarations.
Currently, NRVO produces ideal IR for something like auto r = super.foo(); return r; - the sret pointer is simply forwarded to super.foo().
Returning another return value directly (i.e., return super.foo();) currently performs an obsolete copy and possibly even a postblit call.

[JohanEngelen]

Test will follow once merged. ;)

For dmd-testsuite additions, I understand. But for IR tests, could you add them to the PR? (or perhaps it doesn't involve IR tests yet)

[kinke]

There's no need to check the IR for this. It's actually quite hard to test this (properly visiting the CallExp); the testcase in #1450 can test one (current) symptom if it isn't visited.

[dnadlinger]

Seems like Travis was having network connectivity issues again; hope I didn't miss any failures because of that.

[kinke]

Test: ldc-developers/dmd-testsuite@2fb4d56