ocr: carry the repair receipt out with the page (v0.3.34)

SURYA READS ITALIAN AGAIN. Measured on the live install, the same synthetic page
that scored 29 markers all morning:

    MARKERS : 0    (29 baseline)
    ACCENTS : AEEIOUaeeiou (all twelve)
    La societa` ha facolta` di piu` attivita`: percio` e` cosi` che l'unita`
    perche' gia` oltre.

...but it reported `accents_repaired: null` while visibly having fixed the page,
because _page_from_dict carried `unreadable_trace` across the worker boundary and
never carried `accents_repaired`. The fix worked; only its receipt was dropped.

That matters more than a missing field: the text is no longer byte-for-byte what
the server sent, and a caller comparing the two has no way to know why. A silent
fix is nearly as bad as a silent corruption.

tests: +2 (47 in test_ocr_accent_repair.py) for the receipt crossing the boundary
and for staying absent when nothing was repaired.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PxYHWbCAjjjD7A1rhFKtaW

codai/__init__.py

@@ -16,7 +16,7 @@
 
 # Canonical product version for CoderAI — single source of truth. Both the API
 # metadata and the admin web UI read from here.
-__version__ = "0.3.33"
+__version__ = "0.3.34"
 __tagline__ = "Complete orchestration, distribution and escalation of remotizable advanced inference."
 
 # Configure the CUDA caching allocator BEFORE torch is imported anywhere.

codai/ocr/subprocess_engine.py

@@ -210,4 +210,12 @@ class SubprocessOcrEngine(OcrEngine):
         trace = d.get("unreadable_trace")
         if trace:
             page.meta["unreadable_trace"] = trace
+        # What the worker had to rebuild. Carried for the same reason the trace is:
+        # the text is no longer byte-for-byte what the server sent, and a caller
+        # comparing the two needs to know why. Missing this is why the first
+        # successful repair reported `accents_repaired: null` while visibly having
+        # fixed the page -- the fix worked and only its receipt was dropped.
+        repaired = d.get("accents_repaired")
+        if repaired:
+            page.meta["accents_repaired"] = repaired
         return page

tests/test_ocr_accent_repair.py

@@ -480,3 +480,26 @@ def test_decoding_is_per_id_not_bulk():
     src = inspect.getsource(W._decode_ids)
     assert "for i in ids:" in src
     assert "tok.decode([i]" in src
+
+
+def test_the_repair_receipt_reaches_the_response():
+    """The first successful repair reported `accents_repaired: null` while having
+    visibly fixed the page: the worker emitted the key and _page_from_dict never
+    carried it across. A silent fix is nearly as bad as a silent corruption —
+    nobody can tell the text was rewritten."""
+    from codai.ocr.subprocess_engine import SubprocessOcrEngine
+    page = SubprocessOcrEngine._page_from_dict({
+        "index": 0, "text": "società", "width": 10, "height": 10,
+        "lines": [], "regions": [], "tables": [],
+        "accents_repaired": {"blocks": 1, "recovered": 29, "how": "…"},
+    })
+    assert page.meta["accents_repaired"]["recovered"] == 29
+
+
+def test_a_page_with_nothing_repaired_carries_no_receipt():
+    from codai.ocr.subprocess_engine import SubprocessOcrEngine
+    page = SubprocessOcrEngine._page_from_dict({
+        "index": 0, "text": "ok", "width": 1, "height": 1,
+        "lines": [], "regions": [], "tables": [],
+    })
+    assert "accents_repaired" not in page.meta