Unified JPEG encoder API surface across mozjpeg-rs and zenjpeg

[lilith]

Summary

Make it trivial to swap between mozjpeg-rs and zenjpeg by unifying the native API surface. Both crates should support a common set of builder methods and encode calls so users change only the use statement. zencodec traits should be implemented directly on the native types (feature-gated), not via adapter wrappers.

Target API (works with either crate)

use mozjpeg_rs::{Encoder, Subsampling};  // OR: use zenjpeg::{Encoder, Subsampling};

let jpeg = Encoder::new()
    .quality(85)
    .progressive(true)
    .subsampling(Subsampling::S420)
    .optimize_huffman(true)
    .deringing(true)
    .icc_profile(icc_bytes)
    .exif_data(exif_bytes)
    .encode_rgb(&pixels, width, height)?;

Encoder::max_compression().encode_rgb(&px, w, h)?;
Encoder::fastest().encode_gray(&gray, w, h)?;

With zencodec feature, the native type IS the zencodec config — no adapter wrapper:

// Encoder directly implements zencodec::EncoderConfig when feature="zencodec"
fn encode_generic<C: zencodec::encode::EncoderConfig>(config: C, pixels: PixelSlice<'_>) -> Vec<u8> {
    config.job().encoder().unwrap().encode(pixels).unwrap().into_vec()
}

Three-way API comparison

Aspect	mozjpeg (C wrapper)	mozjpeg-rs (current)	zenjpeg (current)	Unified
Type name	Compress	Encoder	EncoderConfig	Encoder
Constructor	new(ColorSpace)	new(Preset)	ycbcr(q, ss)	new() + explicit ctors
Quality	set_quality(f32)	.quality(u8)	.quality(impl Into<Quality>)	.quality(impl Into<Quality>)
Progressive	set_progressive_mode()	.progressive(bool)	.progressive(impl Into<…>)	.progressive(impl Into<…>)
Subsampling	tuples	Subsampling::S420	ChromaSubsampling::Quarter	Subsampling::S420
Deringing	N/A	.overshoot_deringing(bool)	.deringing(bool)	.deringing(bool)
Force baseline	N/A	.force_baseline(bool)	.force_baseline()	.force_baseline()
ICC	post-start	.icc_profile(Vec<u8>) builder	request().icc_profile(&[u8])	.icc_profile(Vec<u8>) builder
Encode	start+write+finish	.encode_rgb(&[u8], w, h)	.encode_bytes(&[u8], w, h, layout)	.encode_rgb(&[u8], w, h)
zencodec	N/A	Adapter wrapper	Adapter wrapper	Direct trait impl on native type

Tossups resolved toward zenjpeg's design (more elegant signatures).

Changes needed in zenjpeg

Native API additions

1. Encoder type alias — pub type Encoder = EncoderConfig; + root re-export from lib.rs
2. Subsampling re-export — pub use crate::types::Subsampling; from encoder/mod.rs and lib.rs
3. Default impl — EncoderConfig::default() → ycbcr(85, Quarter).progressive(true).optimize_huffman(true)
4. new() — pub fn new() -> Self { Self::default() }
5. Preset constructors — max_compression() and fastest() delegating to OptimizationPreset
6. .subsampling(Subsampling) — builder method converting via existing From<Subsampling> for ChromaSubsampling
7. encode_rgb() / encode_gray() — thin wrappers around encode_bytes() with Rgb8Srgb / Gray8Srgb
8. .icc_profile(Vec<u8>) / .exif_data(Vec<u8>) — metadata fields on EncoderConfig, wired into request() and one-shot encode
9. Un-hide optimize_huffman() if currently #[doc(hidden)]

zencodec: direct trait impl, remove wrapper

Replace JpegEncoderConfig adapter with impl zencodec::EncoderConfig for EncoderConfig directly (feature-gated). The generic_quality_input: Option<f32> field moves onto EncoderConfig. JpegEncodeJob and JpegEncoder remain as thin feature-gated types (they hold per-operation state). Add pub type JpegEncoderConfig = EncoderConfig; for backwards compat.

Corresponding changes in mozjpeg-rs

Will be tracked in imazen/mozjpeg-rs separately:

• Encoder::new() no-arg (old new(Preset) → with_preset(Preset))
• Rename overshoot_deringing → deringing (deprecated alias)
• Add no-arg force_baseline()
• zencodec: impl EncoderConfig for Encoder directly, remove MozjpegEncoderConfig wrapper

Intentional differences (not unified)

	mozjpeg-rs	zenjpeg
Quality scale	mozjpeg u8	jpegli f32 (different tables)
Subsampling::Gray	Exists	Use encode_gray()
Error type	mozjpeg_rs::Error	zenjpeg::Error
Extended features	C compat, smoothing	XYB, AQ, parallel, hybrid trellis, f32/16-bit input