Quantized GGUF style (#1523)

* Metal quantized modifications proposal.

- Add a device param, wherever needed.
- Create new QMetal storage thing that implements QuantizedType.
- Update everywhere needed.

Fix Python.

Fixing examples.

Fix: fmt + clippy + stub.

Moving everything around.

Only missing the actual implems.

Fixing everything + adding dequantized kernels.

More work.

Fixing matmul.

Fmt + Clippy

Some clippy fixes.

Working state.

Q2K Metal -> Bugged (also present in GGML).
Q4K CPU -> Bugged (present previously, new test catch it).
Q5K CPU -> Bugged (present previously).
Q8_1 Both -> Never really implemented it seems
Q8K metal -> Never implemented in metal

Fixing Q2K bug (present in ggml).

* Cleanup.

* Fix the rebase.

* Removing the fences speeds everything up and *is* correct this time...

* Cleanup the fence.

* After rebase.

* Bad code removal.

* Rebase after phi2 merge + fix replit default to CPU.

* Making the CI happy.

* More happy tests.

---------

Co-authored-by: Nicolas Patry <nicolas@Nicolass-MacBook-Pro.local>

candle-examples/examples/blip/main.rs

@@ -106,17 +106,17 @@ pub fn main() -> anyhow::Result<()> {
 
     let config = blip::Config::image_captioning_large();
 
+    let device = candle_examples::device(args.cpu)?;
     let (image_embeds, device, mut model) = if args.quantized {
         let device = Device::Cpu;
         let image = load_image(args.image)?.to_device(&device)?;
         println!("loaded image {image:?}");
 
-        let vb = quantized_blip::VarBuilder::from_gguf(model_file)?;
+        let vb = quantized_blip::VarBuilder::from_gguf(model_file, &device)?;
         let model = quantized_blip::BlipForConditionalGeneration::new(&config, vb)?;
         let image_embeds = image.unsqueeze(0)?.apply(model.vision_model())?;
         (image_embeds, device, Model::Q(model))
     } else {
-        let device = candle_examples::device(args.cpu)?;
         let image = load_image(args.image)?.to_device(&device)?;
         println!("loaded image {image:?}");