AI progress: honest milestone bar from closed JSON contract units

Replace the char-count estimate (EST_OUTPUT_CHARS) with ContractProgress:
points are earned only when real output units complete — first token (5),
each closed contract key / array item (5-90), root closed (95), stored (100).
Reasoning/thinking streams no longer fake progress: the first thinking
delta fires the first-token milestone ('the model is alive') and holds.
Surface the job's tone through ProcessingJobOut/JobInfo so the bar can
label itself 'Summarizing dry wit…'. Test rewritten to pin the new
honest-thinking semantics.

backend/shonar/services/ai/ollama.py

@@ -93,7 +93,8 @@ async def _collect_reply(resp: httpx.Response, ticker) -> str:
 
     Streaming answers are NDJSON (one JSON object per line, ``done`` on the
     last); a server honoring stream=false returns one JSON body — both work.
-    The running character count feeds [ticker] for UI progress."""
+    The accumulated content *prefix* feeds [ticker], which derives progress
+    from completed JSON contract units only (see ContractProgress)."""
     import json as _json
 
     ctype = resp.headers.get("content-type", "")
@@ -103,11 +104,10 @@ async def _collect_reply(resp: httpx.Response, ticker) -> str:
             content = _json.loads(body)["message"]["content"]
         except (ValueError, KeyError, TypeError) as e:
             raise ProviderTransientError("Ollama sent an unreadable reply.") from e
-        ticker(len(content or ""))
+        ticker(content or "")
         return content or ""
 
     parts: list[str] = []
-    total = 0
     async for line in resp.aiter_lines():
         line = line.strip()
         if not line:
@@ -117,14 +117,15 @@ async def _collect_reply(resp: httpx.Response, ticker) -> str:
         except ValueError:
             continue
         piece = (obj.get("message") or {}).get("content") or ""
-        # Like the openai_compat adapter: a thinking channel is real
-        # work and must move the progress ticker even when the server
-        # ignored think:false.
+        # Like the openai_compat adapter: a thinking channel means the
+        # model started (the ticker's first-token milestone) but is not
+        # contract output, so it moves nothing beyond that.
         thinking = (obj.get("message") or {}).get("thinking") or ""
-        if piece or thinking:
+        if piece:
             parts.append(piece)
-            total += len(piece) + len(thinking)
-            ticker(total)
+            ticker("".join(parts))
+        elif thinking:
+            ticker("")
         if obj.get("done"):
             break
     content = "".join(parts)