Support Q8_0 and quantized KV cache types

Currently the library only supports:
```cpp
typedef enum {
    TRANSCRIBE_KV_TYPE_AUTO = 0,
    TRANSCRIBE_KV_TYPE_F32  = 1,
    TRANSCRIBE_KV_TYPE_F16  = 2,
} transcribe_kv_type;
```

Could be cool to allow for Q8 or Q4 quantization for KV cache in low memory scenarios.

