Move embeddings & WikipediaSimilaritySearchTool from API to FW

[michael-desmond]

Is your feature request related to a problem? Please describe.
The WikipediaSimilaritySearchTool and the related embedding capabilities are very useful when working with smaller LLMs. For example with granite bee its really the only way to use the Wikipedia tool. Granite will eventually get a larger context window but the ability to chunk/embed/retrieve seems generally useful for tool developers, and provides some core building blocks for optimized tools.

Right now if I want to provide a granite bee + wikipedia example in the FW I cant use the WikipediaSimilaritySearchTool.

Describe the solution you'd like
Move embeddings & WikipediaSimilaritySearchTool from API to FW to make the functionality available at lower level in stack.

Describe alternatives you've considered
Summarization of wikipedia article using an llm post processing step.

Additional context
None.

[Tomas2D]

We would need to extend the BaseLLM class to introduce the embeddings by adding a new method (similar to tokenize). This can be done in a separate issue. It would do that instead of bringing new classes in.
Regarding WikipediaSimilaritySearchTool, I would create a general tool instead. That tool would wrap up the provided tool, and that's it.

[Tomas2D]

Such a tool can now be created by using the pipe/extend method available on a tool instance.
Example: https://github.com/i-am-bee/bee-agent-framework/blob/main/examples/tools/custom/piping.ts (v0.0.41)

However the functionality is still blocked by #176

[Tomas2D]

Embedding has been added and functionality has been refactored in the API.