confirm-overlap-reject is very general

Minor complaint from looking through the tests and the spec once more.

http://dev.w3.org/webfonts/WOFF2/spec/#conform-overlap-reject says nothing
shall overlap or extend off the end of the file. The surrounding statements
require no more than 3 bytes of extraneous data. It's exactly how WOFF1 did
it, but it seems like it might be overly general (particularly compared to
something like http://www.w3.org/TR/WOFF/#Metadata, which is *laced* with
assertions).

If I put on my test writing hat, having more granular assertions seems like
it would better align with the tests one would write. For example,
TableDirectory has to follow the WOFF2Header (in fact it MUST be at +48),
there can't be a huge hole in between (covered by must start at +48 in this
case), etc.

Thoughts?