The compiler needs to be able to inline the call to compare for the
optimizer to do its job here. When properly inlined, the optimizer can
elide the call to memcmp and do a hardcoded comparison both for the size
and the contents of the string.
It dramatically simplifies the generated code for things like x == ""
and x == "012345678".
Improves the BM_StringRelationalLiteral_Eq_Small_* benchmarks.