Fixed passwords getting percent_encoded

[Thomspoon]

When testing mysql database connectors, I came across this squirrelly response:

thread 'main' panicked at 'called Result::unwrap()on anErr value: InvalidPort', /Users/rustbuild/src/rust-buildbot/slave/stable-dist-rustc-mac/build/src/libcore/result.rs:86

My connection string was similar to:

mysql://root:pass#word@example.com

Which immediately prompted me to wonder what was happening. I cloned the repo for the connector I was using and started debugging, before I dug deeper and found it was due to the url::Url type panicking. So I delved even deeper and ended up here.

I found that parse_userinfo was breaking at the #, which would then cause some errors downstream when parsing the rest of the url.

After this fix, I found that the password, since it wasn't expecting these characters, was percent_encoding these characters, so the connection string would turn into:

mysql://root:pass%23word@example.com

My proposed change would allow additional special characters in the password, accepting the ? and # characters that were previously disallowed. IBM and various others promote the usage of specifically # and ? in passwords, which is why I think those restrictions should be taken out.

This is my first pull request, so please let me know if this was too much.

---

This change is 

[Thomspoon]

On second though, this only works for my use-case, not the main use case of this crate probably. All of the SQL connector repos that I have encountered use the URL crate, so I'm kind of wondering what the workaround for this would be. Url::parse_no_percent_encode?

Sorry for the close and open, I don't know my own strength.

[SimonSapin]

# is not simply disallowed as an arbitrary restriction, it is a separator for the "fragment" component of an URL. For example, structs is the fragment of https://docs.rs/url/1.4.0/url/#structs. As you can see on Travis-CI, your change makes some tests fail.

Using percent-encoding is the correct way.

[Thomspoon]

Ah, that makes sense.

I'm aware why the tests fail, the question now is, would a better way around this problem be to have a parse_without_encoding function? The url crate is such a strong crate, that it wouldn't make sense to have repos such as rust-mysql-simple change their codebase when they could change one function to have the same effect. I'm sure a lot of crates could benefit from that if they are using their own scheme.