A retrieved DOCX needs bounded local text extraction

A file-reading MCP tool returned a staged binary download descriptor, so a successful retrieval did not make Word contents readable to the agent. Fetch the provider-issued URL with restricted hosts, redirects, cancellation and size limits, validate ZIP expansion, then use a DOCX text parser and return plain content while preserving download metadata. Treat extraction failure as a read error rather than green binary-only success; test paragraphs, tables, Unicode, malformed archives and the actual target runtime.