fix(async): chiamate bloccanti in asyncio.to_thread (#36 punto 1)

kaleido (pio.to_image), WeasyPrint (write_pdf), pdfplumber (extract_text_from_pdf)
e la scrittura del CSV di produzione giravano sincrone dentro handler async:
un PDF pesante o un run grande fermava il worker per tutte le altre richieste.
Ora avvolte in await asyncio.to_thread(...) in report_service.py,
ai_service.py e production_export_service.py.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

src/backend/services/production_export_service.py

@@ -9,6 +9,7 @@ The CSV uses the delimiters configured in system_settings, the same ones the man
 export honours, so a shop floor set to the Italian convention (';' and ',') gets
 files that open correctly in their Excel.
 """
+import asyncio
 import csv
 import io
 import logging
@@ -192,7 +193,11 @@ async def export_run_statistics(db: AsyncSession, run: ProductionRun) -> str | N
 
     stamp = (run.closed_at or datetime.now()).strftime("%Y%m%d_%H%M%S")
     filename = f"production_{run.id}_{stamp}.csv"
-    (directory / filename).write_text(content, encoding="utf-8-sig")
+    # A large run's file can take a moment to write: off the event loop, so
+    # closing one run does not stall every other request being served.
+    await asyncio.to_thread(
+        (directory / filename).write_text, content, encoding="utf-8-sig"
+    )
 
     # utf-8-sig: the BOM is what makes Excel read accented characters correctly,
     # and these files are opened in Excel on the shop floor.