Tile op for onnx support

Recently doing SNAC model to rust, the onnx can not be run due to Tile op.

```
Exploring on onnx model: snac_24hkz_sim.onnx
╭───────────────────────────── snac_24hkz_sim.onnx Summary ──────────────────────────────╮
│ IR Version: 7 │
│ Opset Version: 14, │
│ Doc: │
│ Producer Name: pytorch │
│ All Ops: │
│ Gather,Transpose,Conv,Unsqueeze,Tile,Shape,Slice,Concat,Reshape,Add,Mul,Sin,Pow,ConvTr │
│ anspose,ConstantOfShape,RandomNormalLike,Tanh │
╰────────────────────────────────────────────────────────────────────────────────────────╯
snac_24hkz_sim.onnx Detail
╭────────────────────┬────────────────────────┬─────────────────────────┬────────────────╮
│ Name │ Shape │ Input/Output │ Dtype │
├────────────────────┼────────────────────────┼─────────────────────────┼────────────────┤
│ c1 │ [-1, -1] │ input │ int64 │
│ c2 │ [-1, -1] │ input │ int64 │
│ c3 │ [-1, -1] │ input │ int64 │
│ audio_hat │ [-1, 1, -1] │ output │ float32 │
╰────────────────────┴────────────────────────┴─────────────────────────┴────────────────╯
```

any way to adding support for it?

Also, ConvTranspose, RandomNormalLike also useful.

SNAC decoder is now widely used in many TTS and Omni models with Audio speech ability.

Start by locating Candle's ONNX operator import and existing operator implementations, then inspect how the reported snac_24hkz_sim.onnx model reaches the unsupported Tile operation. Reproduce the failure with that model and determine the required coverage for Tile; ConvTranspose and RandomNormalLike are additional requested operators. Done means the model runs past these unsupported operations with appropriate tests.

Written by the indexing model from the issue text.