Adapt frontend/search extraction to new /extract endpoint, per-service env files, ignore tempdocs

Search service now owns document text extraction (PDF/DOCX/PPTX/XLSX) via a
new /extract endpoint, so the frontend proxies to it instead of duplicating
the parsing logic. serve.sh now loads each service's own .env file instead of
a single shared .env. tempdocs/ (transient uploaded query documents) is now
gitignored.