Fix invalid default method

[snapshotpl]

Currently, when method is not provided it's set as empty string. It's invalid state, because it's not possible to set new method (using withMethod) as empty string. It's also other possible invalid state - we can set new method as null, which is not compatible with RequestInterface.

[weierophinney]

This is a slight BC break in that toString() can now raise an exception. As such, I'm going to push it to the next minor release.

[weierophinney]

Hi, @snapshotpl!

I've been reviewing this in a bit more detail, and I've decided to push it to a future 2.0 release.

The reason I'm pushing it to a new major release is because it breaks existing assumptions codified in the tests, particularly around the fact that we previously allowed a client-side request to have a nullable request method; changing this assumption, while a net-positive, causes potential breakage for existing users who may have depended on this "feature".

(Interestingly, PSR-7 does not indicate that there MUST be a value present for the HTTP method, only that getMethod() must return a string. The current implementation does that, except in one case — we currently allow null, which we should not. More generally, PSR-7 does not dictate that a given instance MUST have a valid state, and there are a variety of ways to create an instance with invalid state, including omitting the URI!)

[weierophinney]

Merged with #302.