Add provider-specific prompt caching support in owrap.ai.js (OpenAI + Anthropic)

[nmaguiar]

This PR extends GPT provider handling in js/owrap.ai.js to properly surface prompt-cache token accounting and enable Anthropic prompt caching controls without changing default behavior. Gemini/Ollama behavior remains unchanged aside from documentation clarity.\n\n- OpenAI: capture cache-aware usage stats\n - Extended _captureStats to include nested usage fields when present:\n - usage.prompt_tokens_details.cached_tokens → tokens.cached\n - usage.prompt_tokens_details.audio_tokens → tokens.audio\n - usage.completion_tokens_details.reasoning_tokens → tokens.reasoning\n\n- Anthropic: opt-in prompt caching support\n - Added aOptions.promptCaching (default false).\n - When enabled, both _request and _requestStream send:\n - anthropic-beta: prompt-caching-2024-07-31\n - Extended _captureStats with:\n - usage.cache_creation_input_tokens → tokens.cacheCreation\n - usage.cache_read_input_tokens → tokens.cacheRead\n\n- Anthropic: cache boundary hints in payload\n - When promptCaching is enabled:\n - system is emitted as content blocks with cache_control: { type: \"ephemeral\" }.\n - The last cacheable user message/block is marked with cache_control: { type: \"ephemeral\" } (while avoiding tool_result-only blocks).\n\n- ODoc updates (ow.ai.gpt + ``)\n - Documented promptCaching option for Anthropic.\n - Documented that `getLastStats()` now includes OpenAI cached-token accounting and Anthropic cache read/creation token counters.\n - Added note that Gemini may implicitly cache large system instructions.\n\n- Focused AI tests\n - Added tests for:\n - OpenAI cached/audio/reasoning token capture.\n - Anthropic prompt-caching beta header behavior.\n - Anthropic `cache_control` payload shaping and cache stats extraction.\n\njavascript\nif (isMap(aResponse.usage.prompt_tokens_details)) {\n if (isDef(aResponse.usage.prompt_tokens_details.cached_tokens))\n tokens.cached = aResponse.usage.prompt_tokens_details.cached_tokens;\n}\nif (isMap(aResponse.usage.completion_tokens_details)) {\n if (isDef(aResponse.usage.completion_tokens_details.reasoning_tokens))\n tokens.reasoning = aResponse.usage.completion_tokens_details.reasoning_tokens;\n}\n