harden: per-item robustness + deadlines/caps across connectors; layered tests

Mail: _safe_decode + _iter_messages (per-mail try/except) -> eine kaputte Mail
(charset x-unknown) bricht die Suche nicht mehr ab, findet so auch uralte Mails.
Vollscan + Datum-Sort + MAX_SCAN/DEADLINE_S (Selbstheilung, kein 24h-Haenger mehr).

Files: search_files mit SEARCH_DEADLINE_S + SEARCH_MAX_DIRS begrenzt (war
unbegrenzte rekursive PROPFIND -> real 35s/449 Calls).

Calendar/Contacts: PARSE_DEADLINE_S fuer client-seitiges vobject-Parsen;
MAX_RESULTS-Output-Cap in get_events/get_tasks (war 10939 Zeilen bei 6000 Events);
search Top-N + Hinweise.

Tests: in Schichten getrennt -> Smoke (nightly), Edge + Stress (on-demand).
Neu: test_smoke/test_edge/test_stress + conftest-Marker + run_full_tests.sh,
nightly laeuft nur Smoke. Kontakt-Foto-Roundtrip-Test, Mail-Edge-Mails
(x-unknown, kaputter Header, 2009er Alt-Mail).

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

files/server.py

@@ -5,6 +5,7 @@ from xml.etree import ElementTree as ET
 from typing import Annotated
 
 import httpx
+import time
 from pydantic import Field
 from mcp.server.fastmcp import FastMCP
 from mcp.types import TextContent, ImageContent, EmbeddedResource, BlobResourceContents
@@ -145,18 +146,31 @@ def file_info(
     return "\n".join(parts)
 
 
+SEARCH_DEADLINE_S = 20   # Wall-Clock-Limit fuer search_files
+SEARCH_MAX_DIRS = 400    # max Verzeichnisse (PROPFINDs) pro Suche
+
+
 @mcp.tool()
 def search_files(
     query: Annotated[str, Field(description="Search term — matches file names. Example: 'Rechnung', '.pdf', 'backup'")],
     path: Annotated[str, Field(description="Start directory for search")] = "/",
 ) -> str:
-    """Search for files by name recursively (up to 5 levels deep, max 50 results)."""
+    """Search for files by name recursively (max 5 Ebenen, 50 Treffer; bricht nach
+    SEARCH_MAX_DIRS Ordnern bzw. SEARCH_DEADLINE_S ab, damit der Server nie haengt).
+    Bei grossem Baum/vielen Treffern: 'path' enger setzen."""
     user = get_current_user()
     if not user: return "Error: not authenticated"
     q = query.lower()
     results = []
+    state = {"dirs": 0, "truncated": False}
+    t0 = time.monotonic()
     def _s(p, d=0):
-        if d > 5 or len(results) >= 50: return
+        if d > 5 or len(results) >= 50 or state["truncated"]:
+            return
+        if state["dirs"] >= SEARCH_MAX_DIRS or (time.monotonic() - t0) > SEARCH_DEADLINE_S:
+            state["truncated"] = True
+            return
+        state["dirs"] += 1
         xml, _ = _propfind(user, p)
         for e in _parse_pf(xml, user)[1:]:
             if q in e["name"].lower(): results.append(e)
@@ -166,7 +180,10 @@ def search_files(
     for e in results:
         prefix = "[DIR]" if e["is_dir"] else f"({e['size']:,}b)"
         lines.append(f"{prefix} {e['path']}")
-    return "\n".join(lines) if lines else "Keine Dateien gefunden"
+    out = "\n".join(lines) if lines else "Keine Dateien gefunden"
+    if state["truncated"]:
+        out += f"\n\n[Suche begrenzt nach {state['dirs']} Ordnern/{SEARCH_DEADLINE_S}s — Begriff praezisieren oder 'path' enger setzen]"
+    return out
 
 
 @mcp.tool()