Suggested changes for predictors

[kylebarron]

• Remove registry
• Rename "RevPredict" to Unpredict
• Change to pub(crate) fields
• Remove unused sample_format field
• Remove lifetime on PredictorInfo and Tile and store a single element for bits_per_pixel
• Remove now-unused planar configuration
• Move PredictorInfo into predictor.rs
• Remove unnecessary doctests and unused code (I don't think we need to have doctests for internal APIs like this. If you want to bring back those two tests, let's put them in the tests at the bottom of the file)
• Ensure no copies when data endianness matches system endianness

[feefladder]

Planar configuration is actually important, but I overlooked it in the PR earlier:
output_row_stride<-bits_per_pixel is determined (also) by planar configuration like.

pub fn bits_per_pixel(&self) -> usize {
  match self.planar_configuration {
    Chunky => self.bits_per_sample * self.samples_per_pixel,
    Planar => self.bits_per_sample
  }
}

Then, I don't know exactly if we want to incorporate, and that's a little foggy (tiff2 gave errors on test tiffs, which I think had to do with improper buffer sizing), but I thought something like this was also possible:

/// chunk height (number of rows)
/// If PlanarConfiguration == Planar, we do smart things
pub fn chunk_height_pixels(&self, y: u32) -> usize {
  let res = todo!() // old calculations
  match self.planar_configuration {
    Chunky => res,
    Planar => res * self.samples_per_pixel
  }
}

I'll see if I can get this PR-on-a-PR thing figured out

[feefladder]

all the rest, great changes!

* Remove unnecessary doctests and unused code (I don't think we need to have doctests for internal APIs like this. If you want to bring back those two tests, let's put them in the tests at the bottom of the file)

I already made a test for the doctest, so no problem there!