transform: do not special-case zero or one implementations of a method call

This is a common case, but it also complicates the code. Removing this
special case does have a negative effect on code size in rare cases, but
I don't think it's worth keeping around (and possibly causing bugs) for
such uncommon cases.

This should not result in functional changes, although the output (as
stated above) sometimes changes a little bit.

transform/testdata/interface.out.ll

@@ -47,8 +47,8 @@ typeswitch.notUnmatched:                          ; preds = %0
   br i1 %typeassert.ok, label %typeswitch.Doubler, label %typeswitch.notDoubler
 
 typeswitch.Doubler:                               ; preds = %typeswitch.notUnmatched
-  %doubler.result = call i32 @"(Number).Double$invoke"(i8* %value, i8* null)
-  call void @runtime.printint32(i32 %doubler.result)
+  %1 = call i32 @"(Doubler).Double"(i8* %value, i8* null, i32 %typecode, i8* null)
+  call void @runtime.printint32(i32 %1)
   ret void
 
 typeswitch.notDoubler:                            ; preds = %typeswitch.notUnmatched
@@ -76,6 +76,20 @@ define i32 @"(Number).Double$invoke"(i8* %receiverPtr, i8* %parentHandle) {
   ret i32 %ret
 }
 
+define internal i32 @"(Doubler).Double"(i8* %0, i8* %1, i32 %actualType, i8* %parentHandle) unnamed_addr {
+entry:
+  switch i32 %actualType, label %default [
+    i32 68, label %"reflect/types.type:named:Number"
+  ]
+
+default:                                          ; preds = %entry
+  unreachable
+
+"reflect/types.type:named:Number":                ; preds = %entry
+  %2 = call i32 @"(Number).Double$invoke"(i8* %0, i8* %1)
+  ret i32 %2
+}
+
 define internal i1 @"Doubler$typeassert"(i32 %actualType) unnamed_addr {
 entry:
   switch i32 %actualType, label %else [