chat: cap retrieve_documents per turn, truncate displayed citations, broaden lock-file skip

- MAX_RETRIEVALS_PER_TURN (5): after five retrieve_documents calls in a single
  turn, further calls return a budget-exhausted message instead of executing.
  Caps cost on runaway multi-query loops without forbidding compound questions.

- MAX_CITED_SOURCES (5): accumulated_sources was growing to 14+ entries across
  multiple tool calls and showing chunks Claude never actually used. Cap the
  list returned to the UI at 5, preserving insertion order so the
  highest-relevance early-call results survive. Proper fix (Claude-driven
  inline citations) is bigger work, noted for later.

- ingest.py lock-file skip: changed prefix tuple from ("~$", ".") to ("~", ".")
  so it catches Office lock files even when Nextcloud's filesystem encoding has
  mangled the "$" into a unicode replacement char. Matches what watcher.py
  already does.

scripts/ingest.py

@@ -82,7 +82,9 @@ IGNORED_TOP_FOLDERS = {"Drafts"}
 
 def _ingest_one(filepath: Path, embedder, root: Path = None) -> int:
     """Ingest a single file. Returns chunk count, 0 on skip/failure."""
-    if filepath.name.startswith(("~$", ".")):
+    # "~" catches Office lock files (~$) including the case where Nextcloud
+    # filesystem encoding has mangled the "$" to a unicode replacement char.
+    if filepath.name.startswith(("~", ".")):
         return 0
     if filepath.suffix.lower() not in SUPPORTED:
         return 0