workaround gcc oddity with Future::semi()

Summary:
Prior to this diff, the two `Future::semi()` overloads were logically the same but used different symbols:
* rvalue-qualified overload returned `Future<T>`
* lvalue-qualified overload returned `auto` (which resolved to `Future<T>`)

This seemingly innocuous difference caused gcc to choke: when the `this` object was a `Future<...>` object returned by value, gcc reported the following error: "error: call of overloaded ‘semi()’ is ambiguous", specifically pointing to the two `Future::semi()` overloads.

Conversely clang had no problem with that callsite.

gcc stopped issuing the spurious error-message after this diff's changes, that is, after making the signatures of the two overloads identical in every respect except for the rvalue-qualification vs. lvalue-qualification.

Reviewed By: LeeHowes

Differential Revision: D9475854

fbshipit-source-id: 5fb20f6c57eaf6a5f9b7cb6cdb49782e40704953

folly/futures/Future.h

@@ -1880,7 +1880,7 @@ class Future : private futures::detail::FutureBase<T> {
     return SemiFuture<T>{std::move(*this)};
   }
 
-  auto semi() & {
+  SemiFuture<T> semi() & {
     return std::move(*this).semi();
   }