Add Context-Enabled Semantic Caching recipe to semantic cache folder

[redis-phil]

Add Context-Enabled Semantic Caching Recipe
Overview
This PR adds a comprehensive Jupyter notebook demonstrating Context-Enabled Semantic Caching (CESC) - a three-tier caching architecture that combines Redis vector search with user personalization to deliver tailored LLM responses at 60-80% lower cost and sub-100ms latency.

What's Added

• Complete production-ready implementation using Redis, OpenAI/Azure OpenAI, and comprehensive telemetry tracking

• Real-world IT support scenario showcasing semantic similarity matching with user memory integration for personalized responses

Enterprise Impact
Provides quantifiable ROI with 200-400% first-year returns for organizations processing >10K daily LLM queries through intelligent model selection (GPT-4o for new content, GPT-4o-mini for personalization) and Redis-powered vector caching.

[abrookins]

@redis-phil Thanks for the PR and 👋. Can you take a look at the failing build and make any changes to the notebook needed to get it passing? https://github.com/redis-developer/redis-ai-resources/actions/runs/16683036159/job/47230595099?pr=99

[abrookins]

Hey @redis-phil! We do want recipes to be runnable outside of Colab, ideally. You can check the other recipes for examples of how this usually works (installing redis, using environment variables for secrets, etc.).

A few notes:

• Remove broken AzureLLMClient.query() method - references self.generate_embedding, self.search_cache, self.user_context, self.telemetry, self.llm_client, self.store_response, which don't exist on that class. Seems like dead code?

• Fix vector score logic - Code uses score >= 0.85 but RedisVL with HNSW+cosine returns distance (lower=better) not similarity (higher=better). Need to flip to score <= threshold with different threshold value. Also: why not RangeQuery with threshold? Finally, there is an option, normalize_vector_distance on queries that can flip this from a distance to a similarity value.

• Look into telemetry issues - summarize() prints DataFrame then returns None, which gets printed as "None". Also raw cache hits log latency_ms=0 instead of measuring actual Redis query time.

• You might consider caching more fields - Currently only stores content, content_vector, user_id. Missing prompt (original query), model (which model generated it), created_at, ttl for cache policy management.

• Fix Colab-only secrets handling - Hard imports from google.colab import userdata, crashes outside Colab. Need try/except with fallback to os.getenv(), or just os.getenv().

• Remove unused Redis client - Creates redis_client in cell 5 but never uses it, only uses one via SearchIndex.connect()

• (Style nit) Fix inconsistent naming - GPT4_MODEL vs GPT4mini_MODEL inconsistent, should be something like MODEL_GPT4 and MODEL_GPT4_MINI

• Update Colab badge - Points to private Drive ID that 404s for others. E.g. "<a href=\"https://colab.research.google.com/github/redis-developer/redis-ai-resources/blob/main/python-recipes/vector-search/01_redisvl.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n"