feat(ocr): per-script-type confidence thresholds

Kurrent OCR produces much lower confidence than typewriter/Latin.
Separate thresholds allow aggressive filtering for Kurrent (0.5)
while keeping typewriter lenient (0.3).

- OCR_CONFIDENCE_THRESHOLD: default for Surya paths (0.3)
- OCR_CONFIDENCE_THRESHOLD_KURRENT: Kraken Kurrent path (0.5)
- apply_confidence_markers() now accepts threshold parameter
- get_threshold(script_type) selects the right threshold

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

ocr-service/test_confidence.py

@@ -2,7 +2,7 @@
 
 import os
 import pytest
-from confidence import apply_confidence_markers, words_from_characters
+from confidence import apply_confidence_markers, words_from_characters, get_threshold
 
 
 # ─── apply_confidence_markers ─────────────────────────────────────────────────
@@ -75,22 +75,29 @@ def test_just_below_threshold_replaced():
     assert apply_confidence_markers(words) == "[unleserlich]"
 
 
-def test_custom_threshold_via_env(monkeypatch):
-    monkeypatch.setenv("OCR_CONFIDENCE_THRESHOLD", "0.8")
-    # Need to reload the module to pick up the new env var
-    import importlib
-    import confidence
-    importlib.reload(confidence)
-
+def test_custom_threshold_via_parameter():
     words = [
         {"text": "Lieber", "confidence": 0.95},
         {"text": "Freund", "confidence": 0.5},
     ]
-    assert confidence.apply_confidence_markers(words) == "Lieber [unleserlich]"
+    assert apply_confidence_markers(words, threshold=0.8) == "Lieber [unleserlich]"
+    assert apply_confidence_markers(words, threshold=0.3) == "Lieber Freund"
 
-    # Reset
-    monkeypatch.setenv("OCR_CONFIDENCE_THRESHOLD", "0.3")
-    importlib.reload(confidence)
+
+def test_kurrent_threshold_is_higher_than_default():
+    default = get_threshold("TYPEWRITER")
+    kurrent = get_threshold("HANDWRITING_KURRENT")
+    assert kurrent > default
+
+
+def test_get_threshold_kurrent():
+    assert get_threshold("HANDWRITING_KURRENT") == 0.5
+
+
+def test_get_threshold_default():
+    assert get_threshold("TYPEWRITER") == 0.3
+    assert get_threshold("HANDWRITING_LATIN") == 0.3
+    assert get_threshold("UNKNOWN") == 0.3
 
 
 def test_low_confidence_at_start():