Make Best one network, and retire Medium and the mixture

Best was a mixture of two experts and a gate, 110 GMAC a megapixel;
Medium a single network at 48 that was softer on real edges. fb-combo
(darkroom-denoise, 20 000 steps from fb-edges2, taught by the mixture
with a quarter of its crops from the edge-rich parts of the frames) is
Medium's shape and holds the mixture's edges on real photographs:
edge PSNR within 0.04-0.06 dB at ISO 1600/6400/25600, more sharpness
kept at all three, the chart's edge 0.89 photosites wide against 0.82.
It is 0.27 dB short on smooth areas at ISO 25600. It becomes Best, and
the methods are Bilinear, Fast and Best.

Saved edits keep their numbers: 2, which was Medium, is now Best, and
3, which was Best, is past the end and reads as the default, Best.
The network ships as mosaic-hq, a new name: the result cache keys a
model by name and size, and this one is byte for byte the old Medium's
size. Its tablet form (A16W16) lost 0.00 dB in simulated QDQ at every
ISO and at most 0.09 dB across the noise bracket.

core/dr-denoise/src/lib.rs

@@ -44,19 +44,15 @@ pub const FAST: Shipped = Shipped {
     whole: "mosaic-fast.onnx",
     halo: HALO,
 };
-/// A student of the mixture with the first release's shape: 3.2 M
-/// parameters, 48 GMAC a megapixel.
-pub const MEDIUM: Shipped = Shipped {
-    file: "mosaic-medium-1408.onnx",
-    whole: "mosaic-medium.onnx",
-    halo: HALO,
-};
-/// The mixture: a flat expert, an edge expert and the gate that blends them.
-/// It reaches further than either, so it keeps a smaller centre of each tile.
+/// One network of the first release's shape, 3.2 M parameters and 48 GMAC a
+/// megapixel, taught by the mixture of experts that was Best until 0.24:
+/// its edges at a third of its work (denoise.md §15). A new file name, not
+/// the old Medium's or Best's: the result cache keys a model by its name
+/// and size, and this one is byte for byte the old Medium's size.
 pub const BEST: Shipped = Shipped {
-    file: "mosaic-best-1408.onnx",
-    whole: "mosaic-best.onnx",
-    halo: 256,
+    file: "mosaic-hq-1408.onnx",
+    whole: "mosaic-hq.onnx",
+    halo: HALO,
 };
 
 #[derive(Debug, thiserror::Error)]

core/dr-denoise/src/onnx.rs

@@ -11,11 +11,11 @@
 //! with the Bayer packing spelled `SpaceToDepth`, which QNN can hold and the
 //! 6-D reshape it replaces it cannot.
 //!
-//! Each network also ships with any height and width (`mosaic-best.onnx`
-//! beside `mosaic-best-1408.onnx`, darkroom-denoise `tools/export_whole.py`,
+//! Each network also ships with any height and width (`mosaic-hq.onnx`
+//! beside `mosaic-hq-1408.onnx`, darkroom-denoise `tools/export_whole.py`,
 //! identical to the fixed file at 1408²). Where the rung takes any size, the
-//! frame runs whole instead of in tiles whose borders are thrown away —
-//! Best computes 2.47 photosites for every one it keeps in 1408² tiles
+//! frame runs whole instead of in tiles whose borders are thrown away — a
+//! 1408² tile keeps 1024², 1.89 photosites computed for each one kept
 //! (denoise.md §14).
 
 use crate::tile::{Sizes, TileNet};