feat: use file extensions from collection names for MIME guessing

[Frando]

I was trying to serve a website (HTML and CSS) from an iroh collection with the iroh-gateway example. It works, but CSS is not applied to the page because the gateway serves CSS files with text/plain, which makes the browser ignore it.

The MIME guesser currently only uses the blob data to guess the mime type (via mime_classifier. This is insufficient in many cases (e.g. all text files will be text/plain).

This PR changes this so that for blobs served through collections, the extension of the filename from the collection metadata is used to guess the MIME type. If the file has no or an unknown extension, it falls back to the mime_classifier.

With this change, we can serve websites. At least basic static sites.

[rklaehn]

Looks reasonable. So the extension takes precedent over the content?