use String instead of RequestUri for HTTP path

[pyfisch]

Addresses #8. Do not merge yet.

You said the path could be a slice of a buffer, but using &str with the current code does not work since the Transport in parse_headers is borrowed too long mutably by Head. I think it is possible to use slices but parse_headers needs to be rewritten. Maybe it would be easier if the headers weren't passed to request_start again?

[tailhook]

Hm, the idea was that Head is passed by reference to the headers_received, because rotor still needs some info from headers. Then it passes ownership to the application in request_start.

Perhaps it's possible to extract everything needed before executing headers_received (if it doesn't require to clone/allocate memory). Then we can get rid of request_start entirely.

[pyfisch]

• fn headers_received(head: &Head, scope: &mut Scope<Self::Context>) -> Result<(Self, RecvMode, Deadline), StatusCode>;
• fn request_start(self, head: Head, response: &mut Response, scope: &mut Scope<Self::Context>) -> Option<Self>;

The methods differ in their return type and that request_start expects a response object. If request_start is removed there is no way to construct the response immediatly after the headers were received or should the headers_received method be extended to accept a response object?

[tailhook]

or should the headers_received method be extended to accept a response object?

Yes, exactly. (or at least I don't see any difficulties with that at the moment)

[pyfisch]

There may be a slight increase in memory allocation because the Response is also created for failing requests.

[tailhook]

There may be a slight increase in memory allocation because the Response is also created for failing requests.

No, Response does not have heap allocated parts.