Provide pre-pipeline function to modify options based on input image metadata

Nobody has claimed this yet.

This could seriously increase the power of Stream-based image processing.

Here's an example of how halving an image's dimensions could work:

``` javascript
// PROPOSED API NOT YET AVAILABLE
var halver = sharp().before(function(metadata) {
this.resize(metadata.width / 2, metadata.height / 2);
});
readableStream.pipe(halver).pipe(writableStream);
// PROPOSED API NOT YET AVAILABLE
```

The existing `metadata` logic can also be improved to require only the first few hundred bytes of a Stream.

In addition, adding a "playbook" of example uses for this feature to the docs would be great too. There's a variable/percentage extract/crop under discussion at #205 that this should allow for.

Start by reading the existing metadata logic and the stream-processing entry point. Trace how image metadata becomes available, then define how the proposed before callback would modify options and how metadata could require only the first few hundred stream bytes; done includes the API behavior and documented example uses.

Written by the indexing model from the issue text.