Change Identify API to accept only an io.Reader

[jhwz]

Implements a "headerReader" which caches all reads, and returns an io.MultiReader so we can return a reader which will read from the same point in the input stream.

This is particularly useful for the Identify function, which can now just accept an io.Reader and not require an io.Seeker which is somewhat unnecessary. Also, it will also always return the Reader despite the error which the caller can then read from as normal.

On reflection it may be useful to still support io.Seeker's in another function, maybe these changes should be moved somewhere else. It is more useful for files because you can still continue to use the *os.File instead of the more restrictive io.Reader.

Also includes some tests for the mentioned functions.

This PR is based somewhat off the discussion in #317. I like this package and want to shift some of my projects that have a less clear API to using this. As an alternative, maybe if we expose the formats map, I can just implement this in a separate project.

[mholt]

Thanks for this PR, I've been swamped catching up on things after a bout of sickness, but I like where this is going and I'll try to review it soon!

(A breaking change is OK since we're still in development, pre-release.)

[mholt]

@jhwz Thanks for this change. I took the liberty of refactoring a bit and renaming things to clean it up and make it more consistent with the existing code style. (I also moved the new code into formats.go and formats_test.go instead of their own separate files.) I also fixed a minor bug in the tests. I appreciate the thought that went into your change; I also think we don't have to do as much of the buffer management ourselves though, we can use existing types in the standard library, more like how I did it in Caddy's Layer 4 module: https://github.com/mholt/caddy-l4/blob/56bd7700d889f2ffd52353c241819ddbc7745ff6/layer4/connection.go#L72-L108

I think it'll be easier to maintain and is a little easier to read, and less error-prone.

Anyway, thank you for the contribution! I'm going to resolve the merge conflict and then finish this.

[jhwz]

@mholt Just flicked through your changes - made a lot of sense. Thanks for merging - the new API will be much more convenient for my use cases!

[mholt]

Thanks again for the contribution. Let me know if you have any questions or feedback! 👍