Docs: Add vignette for transfer learning with Keras Applications

[davidrsch]

What part of the documentation is this about?

This is a request for a new piece of documentation, likely a new vignette, on how to perform transfer learning with pre-trained Keras Applications.

What is the issue?

There is currently no documentation explaining how to perform transfer learning using models from keras::applications (e.g., ResNet50, VGG16) within a kerasnip and tidymodels workflow. This is a very common and powerful deep learning pattern, and its absence from the documentation is a significant gap.

Users are left to figure out on their own how to correctly structure the model specification, freeze the weights of the pre-trained base model, and attach a new trainable classification head on top of it.

Suggested improvement

I propose creating a new vignette (e.g., vignettes/applications.Rmd) that provides a complete, end-to-end example of transfer learning. This guide should clearly demonstrate how to:

1. Define a kerasnip layer block that correctly wraps a Keras Application.
2. Show the critical step of freezing the base model's weights for feature extraction by setting application_block$trainable <- FALSE.
3. Use the Keras Functional API via create_keras_functional_spec() to attach a new, trainable head to the frozen base.
4. Integrate the entire model into a standard tidymodels workflow for fitting and evaluation.

Additional context

Providing this documentation would resolve the underlying problem that led to my original feature request for a block_from_application() helper function. A clear, documented example showing the existing, flexible pattern is a much better and more empowering solution than adding a specialized function to the API.