Migrate LLM API cache from pickle to JSON; extract shared LLMCacheBase class

[Copilot]

The API cache stores only dicts, so pickle is unnecessary. The Ollama client duplicated the caching logic without sharing a common base.

Changes

• LLMCacheBase class in openai_client.py — extracts _save_cache, _load_cache, set_api_cache using json.dump/json.load with UTF-8 encoding
• OpenAIClient inherits from LLMCacheBase, removes redundant cache methods and pickle import
• OllamaClient inherits from LLMCacheBase instead of reimplementing caching standalone
• AzureClient unchanged — inherits transitively via OpenAIClient
• Default cache filename changed from openai_api_cache.pickle → openai_api_cache.json in config.ini, examples/config.ini, and __init__.py
• Tests updated from pickle to JSON assertions; added TestLLMCacheBase (6 tests) and TestOllamaClientCache (5 tests) covering inheritance, roundtrip, corruption handling

Inheritance

LLMCacheBase          ← _save_cache, _load_cache, set_api_cache (JSON)
├── OpenAIClient
│   └── AzureClient
└── OllamaClient

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.