16 bit depth images are saved using host endian

[ColonelThirtyTwo]

Test program that writes a 16-bit-depth RGBA image containing a gradient:

extern crate lodepng;
extern crate rgb;
use std::iter::repeat;
use rgb::RGBA;

fn main() {
	let height = 500;
	let row = (0..u16::max_value()).map(|i| RGBA{r: u16::max_value(), g: 0, b: 0, a: u16::max_value()-i}).collect::<Vec<_>>();
	let color_buffer: Vec<RGBA<u16>> = repeat(&row).take(height).flat_map(|row| row).map(|pixel| *pixel).collect::<Vec<_>>();
	let _ = lodepng::encode_file("test.png", color_buffer.as_slice(), u16::max_value() as usize, height, lodepng::ColorType::RGBA, 16);
}

This produces the following image on my x86-64 laptop, which is not a single solid gradient as expected:

If I flip the endianness of the image, I get the correct result:

gm convert test.png -endian LSB -depth 16 RGBA:test.raw
gm convert -endian MSB -depth 16 -size 65535x500 RGBA:test.raw test-flipped.png

This is likely because encode_file doesn't actually use the type of the array its passed; it just converts it to a u8 array and feeds it straight into the encoder. While that seems like a good idea for making the encoder type agnostic, issues like this combined with the fact that rust doesn't guarentee that fields are stored in the order that they are specified make this a bad idea. Since you depend on the rgb crate anyway, you may as well take pixels using the rgb crate's structures.