feat: integrate AlloyDB AI features

[madamsetty-pavan]

Integrate AlloyDB AI Features (Fixes b/531729585)

Description

This pull request brings comprehensive support for the latest AlloyDB AI capabilities directly into the LangChain provider. The integration enables developers to leverage AlloyDB's native Postgres extensions for advanced AI workflows without manually writing raw SQL strings or building ad-hoc wrappers.

Features Implemented

1. RUM Indexing for Full-Text Search (b/531741249)
   • Added the RUMIndex dataclass to indexes.py for advanced text-search index configurations.
2. Multimodal Embeddings (b/531740488)
   • Added embed_image() and aembed_image() methods to AlloyDBEmbeddings.
   • Integrates seamlessly with AlloyDB's google_ml.image_embedding() function.
3. Auto Vector Embeddings (b/531739593)
   • Added initialize_auto_vector_embeddings() to AlloyDBVectorStore (and its async counterpart).
   • Allows automatic invocation of ai.initialize_embeddings() on the target tables.
4. Time-Series Forecasting Integrations (b/531740822)
   • Built forecast() and aforecast() integrations on the AlloyDBEngine class.
   • Interfaces directly with the native google_ml.forecast() table function.
5. AI Functions - Sentiment Analysis & Summarization (b/531739592)
   • Introduced AlloyDBSentimentTool and AlloyDBSummaryTool in a new tools.py module.
   • Provides native LangChain BaseTool abstractions that call google_ml.sentiment_analysis() and google_ml.summarize_content().
6. Columnar Engine Integration (b/531740612)
   • Added enable_columnar_engine() and enable_auto_columnarization() methods to the VectorStore classes to leverage google_columnar_engine_add.
7. Vector Assist Support (b/531740441)
   • Exposed define_vector_assist_spec(), apply_vector_assist_spec(), and get_vector_assist_recommendations() in the VectorStore.
   • Utilizes the vector_assist Postgres extension to optimize vector queries automatically.
8. Natural Language to SQL (b/531740565)
   • Created a native AlloyDBToolkit and AlloyDBNL2SQLTool inside toolkit.py.
   • Delegates Natural-Language-to-SQL generation to AlloyDB's alloydb_ai_nl.get_sql() / google_ml.generate_sql() instead of relying on generic LLM schemas, taking advantage of local model tuning and native Postgres understanding.

Motivation and Context

Closes the gap on b/531729585, bringing langchain-google-alloydb-pg-python up to parity with the latest and greatest AlloyDB AI features.

How Has This Been Tested?

The features hook into existing AlloyDBEngine connectivity and leverage _run_as_sync and _run_as_async patterns established within the codebase. The Python syntax is validated successfully. Next steps for the author would be adding integration tests against a live AlloyDB instance equipped with the google_ml_integration and vector_assist extensions.

[madamsetty-pavan]

Superseded by #621, #622, and #623 to make reviews easier.