feat: import DOCX/PDF/scanned templates via DeepSeek recognition

Backend pipeline:
- POST /api/templates/import (multipart, max 25 MB)
- extract.ts: DOCX→mammoth, PDF→pdf-parse, fallback to OCR via tesseract+poppler-utils
  (pdftoppm renders pages to PNG, tesseract reads with rus+eng)
- deepseek.ts: chat completions client with strict JSON response_format
- recognize.ts: structured prompt that produces simplified DocBody (string text),
  postprocessor wraps text in TipTap-compatible JSON, validates with zod schema
- prompt enforces placeholder substitution: {{customer.*}}, {{executor.*}},
  {{contract.number}}, {{contract.date}}, {{today}}
- error codes: NO_OCR / NO_DEEPSEEK_KEY / UNSUPPORTED_MIME / INVALID_DOC_BODY

Dockerfile: apk add tesseract-ocr (+rus +eng data), poppler-utils, imagemagick

Frontend:
- Templates page: ⤴ Загрузить документ → file picker (.docx,.pdf,.png,.jpg)
- doc type selector (contract/invoice/act/upd)
- import-banner with spinner shows uploading→analyzing stages
- on success navigates to /templates/:id (TemplateEdit) for review

Reuses DEEPSEEK_API_KEY pattern from Hall-planer.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

apps/api/package.json

@@ -20,15 +20,20 @@
     "@fastify/cookie": "^9.4.0",
     "@fastify/cors": "^9.0.1",
     "@fastify/helmet": "^11.1.1",
+    "@fastify/multipart": "^10.0.0",
     "@prisma/client": "^5.22.0",
     "fastify": "^4.28.1",
     "fastify-plugin": "^4.5.1",
     "jose": "^5.9.6",
+    "mammoth": "^1.12.0",
+    "node-tesseract-ocr": "^2.2.1",
+    "pdf-parse": "^2.4.5",
     "puppeteer-core": "^24.42.0",
     "zod": "^3.23.8"
   },
   "devDependencies": {
     "@types/node": "^20.16.0",
+    "@types/pdf-parse": "^1.1.5",
     "embedded-postgres": "^18.3.0-beta.17",
     "pino-pretty": "^11.3.0",
     "prisma": "^5.22.0",