(encoded_)set_path should bypass edit_segments

[alandefreitas]

The library has the convention that "/" (string_view) -> {} (segments) and "/./" (string_view) -> {""} (segments). Functions that go through edit_segments (insert, replace, ...) apply prefixes as required to maintain the Sequence representation. Meanwhile, normalization might not respect the mapping from segments -> Sequence.

url::set_path and url::set_encoded_path, on the other hand, has always assumed it can reuse edit_segments but that's incorrect because it should set the complete path string to exactly what the user asked for instead of attempting to maintain their sequence representation. Not fixing this leads to weird results such as assert(url("").set_path("/./") == "/././").

Thus, set_path should think in terms of the path string and only include extra prefixes where invalid urls would result. On the other hand, edit_segments should only be used when thinking in terms of the sequence: when we want the mapping of the path to the sequence.

Besides these wrong results, making edit_segments do two different things is dangerous, setting the complete path at once is more efficient, and removing workarounds from edit_segments will also make it more efficient.