Add Google GenAI embedding model support

[azanux]

Summary

• Adds support for Google GenAI (Gemini) embedding models
• Introduces gemini-embedding-001 (3072 dimensions) as the recommended GA embedding model
• Uses Spring AI 1.1.1's spring-ai-google-genai-embedding artifact with GoogleGenAiTextEmbeddingModel and GoogleGenAiEmbeddingConnectionDetails
• Supports both API key and Vertex AI authentication modes, consistent with the existing LLM configuration

Documentation

• Added embedding models list, updated example application.yml with default-embedding-model
• Added embedding models table (name, model ID, dimensions, pricing)

Note on createGoogleGenAiEmbeddingConnectionDetails()

Unlike other providers (OpenAI, Bedrock, Ollama) where a single API client is shared between chat and embedding models, Spring AI exposes two distinct connection types for Google GenAI:

• Client for the chat model
• GoogleGenAiEmbeddingConnectionDetails for the embedding model

These two classes are incompatible, requiring a separate factory method (createGoogleGenAiEmbeddingConnectionDetails()) that mirrors the authentication logic (API key vs Vertex AI) from createGoogleGenAiClient(). This is a constraint imposed by the Spring AI library.

Test plan

• Module compiles successfully
• All tests pass (26 existing + 13 new embedding tests)
• Verify in ragbot that Google GenAI embeddings appear in startup logs

Changes

pom.xml

• Added spring-ai-google-genai-embedding dependency (managed by Spring AI BOM)

GoogleGenAiModelLoader.kt

• Added GoogleGenAiEmbeddingModelDefinition and EmbeddingPricingModel data classes
• Extended GoogleGenAiModelDefinitions with embeddingModels property
• Added embedding model validation (name, modelId, dimensions, pricing)

GoogleGenAiModelsConfig.kt

• Added createGoogleGenAiEmbeddingConnectionDetails() with same auth logic as LLM client
• Added createGoogleGenAiEmbedding() that creates a SpringAiEmbeddingService wrapping GoogleGenAiTextEmbeddingModel
• Updated googleGenAiModelsInitializer() to register embedding beans alongside LLMs and return registeredEmbeddings in ProviderInitialization

google-genai-models.yml

• Added embedding_models section with gemini-embedding-001

GoogleGenAiModelLoaderTest.kt

• Added 16 new tests covering embedding model loading, validation (invalid dimensions, blank name/modelId, negative pricing), and combined LLM+embedding scenarios