Changes
- Improve Read Filings tool with structured planning: first LLM call returns ticker, filing types, and limit; filings metadata and item types are fetched in parallel
- Add support for multiple filing types per request in get_filings (filing_type is now an array)
- Add item descriptions to filing item type prompts and cache item types in memory for reuse
- Cap Read Filings at 3 concurrent filing reads and improve error handling and query escaping in prompts
- Improve cache request labels to include all params (e.g. interval, limit) for easier debugging