Add Depth Anything with transformers.js

[nasif-co]

Implements the depth-anything-v2-small model in ml5 using transformers.js. This current draft does the following:

• Maintains compatibility with the existing ml5.depthEstimation user-facing code and examples.
• Keeps the logic in a separate class from the tensorflow model.
• For the time being, imports transformers.js through a script that is appended to the html. Moving forward, when transformers.js gets bundled (in Add Transformer.js as a backend for image classification #289 or another pull request specific to doing just the bundling), this will be replaced.

With this in mind, the pending tasks are:

• Autodetect the ideal device and dtype for the current browser, and set those automatically.
• Deal with the onnxruntime console warnings (mentioned here Add Transformer.js as a backend for image classification #289 (comment) and here [Question] How to stop warning log huggingface/transformers.js#270)
• Replace the transformersLoader script with the bundled transformers.js

[nasif-co]

For ease of use, here are two examples on the p5 web editor using a build of this code:

• DepthAnything Webcam Estimation
• DepthAnything Estimated Mesh

You can swap back to the tensorflow depth estimation on these examples by replacing this:

depthEstimator = await ml5.depthEstimation('depth-anything-v2-small');

with this:

depthEstimator = await ml5.depthEstimation();

This build lacks automatic detection of the best device and dtype, so it is only guaranteed to run on Chromium-based browsers. This is the next issue to be addressed.