Improve lossless compression efficiency

The KTX v2 spec supports vanilla zlib / deflate and Zstandard schemes. On their own, they do not work well with raw pixel data, especially of floating-point formats.

One option to boost compression is to apply a lossy RDO pass to the input data. Another is to transform raw values into something more suitable for subsequent compression steps. An example of the latter is lossless [PNG Filtering](https://www.w3.org/TR/PNG/#9Filters).

We need to design and integrate a reasonable pre-processing solution that could be used for lossless workflows.

Start by reading the KTX v2 specification's lossless compression requirements and the linked PNG Filtering section, then compare the stated zlib/deflate and Zstandard workflows for raw and floating-point pixel data. The issue names no implementation files or tests; done means a reasonable preprocessing approach has been designed and integrated for lossless workflows.

Written by the indexing model from the issue text.