Merge: Basename fix

Should close #1736 if performances are adequate for @privat.

This PR optimizes some methods in both `SequenceRead` and `FlatString`, on the test program exposed in #1736, the valgrind runtime has gone from 418,872,526 Ir to 136,522,095 Ir (~67.5% improvement)

Some more improvements could be observed if we could get rid of some useless Boxes in `NativeString::to_s_with_copy` because of a `is_same_instance` which boxes NativeStrings (24k allocations in total to remove).

Pull-Request: #1740
Reviewed-by: Jean Privat <jean@pryen.org>

diff --cc lib/core/collection/abstract_collection.nit
Simple merge