Split NIFFile into two parts, NIFFile which is cached and is a container
for a parsed NIF, and NIFStream which is a class specialized for parsing
NIFs. This required a semi-sweeping change to make all record classes
accept a NIFStream instead of a NIFFile as an agurment to their read
functions.
Slice arrays use pre-allocated pointers whose memory is managed externally. This is
unnecessary and ultimately detrimental since it prevents any kind of data fixup
(e.g. little endian to big endian, p[adding handling), and it also makes it difficult
to use Ogre data streams.
This moves the index resolution into a separate post method instead of always
checking when access. As a result, it reduces the size of it down to the size
of a pointer, as opposed to 2 pointers + 1 int. The appropriate methods are
added to the various node types to make sure they're resolved.