This PR fixes some small issues that was making examples not run on
macOS.

The code has been validated through the dora-rerun node.
```
nodes:
  - id: dora-pyorbbecksdk
    path: dora-pyorbbecksdk
    inputs:
      tick: dora/timer/millis/30
    outputs:
      - image
      - depth_image
      - depth_data

  - id: view
    path: dora-rerun
    inputs:
      image_1:
        source: dora-pyorbbecksdk/image
        queue_size: 1
      depth_image:
        source: dora-pyorbbecksdk/depth_image
        queue_size: 1
```
![屏幕截图 2025-01-03
143912](https://github.com/user-attachments/assets/2564bb53-6da0-47d1-824f-214f893135f1)

In addition to depth images, I also added output depth data to
facilitate the processing of depth information.

This makes it easier to understand visualization.

## Before

```mermaid
        flowchart TB
  dora-microphone
  dora-vad
  dora-distil-whisper
  dora-rerun[/dora-rerun\]
subgraph ___dora___ [dora]
  subgraph ___timer_timer___ [timer]
    dora/timer/secs/2[\secs/2/]
  end
end
  dora/timer/secs/2 -- tick --> dora-microphone
  dora-microphone -- audio --> dora-vad
  dora-vad -- audio as input --> dora-distil-whisper
  dora-distil-whisper -- text as original_text --> dora-rerun
```

## After

```mermaid
        flowchart TB
  dora-microphone[**dora-microphone**<hr/>*Microphone sending samples of audio data*]
  dora-vad[**dora-vad**<hr/>*Voice Activity Detection using Sidero model filtering audio samples into audio samples including voice.*]
  dora-distil-whisper[**dora-distil-whisper**<hr/>*Whisper model able to transcribe audio.*]
  dora-rerun[/**dora-rerun**<hr/>*Real time visualisation using Rerun.*\]
subgraph ___dora___ [dora]
  subgraph ___timer_timer___ [timer]
    dora/timer/secs/2[\secs/2/]
  end
end
  dora/timer/secs/2 -- tick --> dora-microphone
  dora-microphone -- audio --> dora-vad
  dora-vad -- audio as input --> dora-distil-whisper
  dora-distil-whisper -- text as original_text --> dora-rerun
```

Making installing the cli from pip auto install the api from pip. If
there's already one installed it will skip it, if there's none it's
going to install the latest version.

This PR add all code required to finetune and do inference of RDT 1B on
Agilex Piper.

The current WIP is on making the code cleaner.