feat: implement full-platform STT API (v2.3 consensus plan)

Batch+realtime transcription API: faster-whisper engine w/ EngineOwner
(single GPU owner, OOM downgrade chain, persisted attempted_profiles),
hardware-adaptive device manager (T0-T3 VRAM tiers), Redis/in-proc job
queue w/ leases + crash recovery, postprocess (glossary/rules/LLM w/
egress guard), privacy-first result store (UUID keys, source deleted
after transcribe), retention sweeper, API-key auth (HMAC digests,
scopes, job ownership), WebSocket realtime lane (LocalAgreement),
CLI (detect/transcribe/bench/serve/key), Docker, benchmark runner.

127 mock-based tests pass; ruff clean. Includes verification checklist
and autoplan review notes.

tests/unit/test_metrics.py

@@ -0,0 +1,84 @@
+"""벤치마크 지표 단위 테스트 — WER/K-CER/entity 보존."""
+
+from __future__ import annotations
+
+from luke_scribe.benchmark.metrics import (
+    character_kcer,
+    clip_metrics,
+    entity_retention,
+    normalize_kcer,
+    word_wer,
+)
+
+
+class TestKcer:
+    def test_normalization_nfkc_punct(self):
+        # NFKC: 유사 문자 통일, 문장부호/공백 제거
+        assert (
+            normalize_kcer("오늘 API 서버에서, vLLM을 사용해 보겠습니다.")
+            == "오늘api서버에서vllm을사용해보겠습니다"
+        )
+
+    def test_exact_match_zero(self):
+        assert character_kcer("오늘 API 사용", "오늘 API 사용") == 0.0
+
+    def test_punct_diff_zero(self):
+        # 문장부호 차이는 K-CER에서 0 (정규화됨)
+        assert character_kcer("오늘 API 사용.", "오늘 API 사용") == 0.0
+
+    def test_one_char_diff(self):
+        ref = "오늘 API 사용"
+        hyp = "오늘 APT 사용"
+        assert 0 < character_kcer(ref, hyp) < 0.3
+
+
+class TestWer:
+    def test_exact(self):
+        assert word_wer("a b c", "a b c") == 0.0
+
+    def test_one_sub(self):
+        assert word_wer("a b c", "a x c") == 1 / 3
+
+    def test_empty_ref(self):
+        assert word_wer("", "") == 0.0
+        assert word_wer("", "a") == 1.0
+
+
+class TestEntityRetention:
+    def test_preserved(self):
+        ref = "오늘 API 서버에서 vLLM을 사용합니다"
+        hyp = "오늘 API 서버에서 vLLM을 사용합니다"
+        entities = [
+            {"canonical": "API", "surface": "API", "start_char": 3, "end_char": 6},
+            {"canonical": "vLLM", "surface": "vLLM", "start_char": 12, "end_char": 16},
+        ]
+        p, t = entity_retention(ref, hyp, entities)
+        assert p == 2 and t == 2
+
+    def test_lost(self):
+        ref = "오늘 API 서버"
+        hyp = "오늘 에이피아이 서버"
+        entities = [{"canonical": "API", "surface": "API", "start_char": 3, "end_char": 6}]
+        p, t = entity_retention(ref, hyp, entities)
+        assert p == 0 and t == 1
+
+    def test_one_occurrence_one_match(self):
+        # 같은 entity가 두 번 등장 → 2개 annotation, 둘 다 보존
+        ref = "API 서버와 API 게이트웨이"
+        hyp = "API 서버와 API 게이트웨이"
+        entities = [
+            {"canonical": "API", "surface": "API", "start_char": 0, "end_char": 3},
+            {"canonical": "API", "surface": "API", "start_char": 8, "end_char": 11},
+        ]
+        p, t = entity_retention(ref, hyp, entities)
+        assert p == 2 and t == 2
+
+
+class TestClipMetrics:
+    def test_perfect(self):
+        ref = "오늘 API 서버에서 vLLM을 사용합니다"
+        m = clip_metrics(
+            ref, ref, [{"canonical": "API", "surface": "API", "start_char": 3, "end_char": 6}]
+        )
+        assert m.wer == 0.0 and m.k_cer == 0.0
+        assert m.entities_preserved == 1 and m.entities_total == 1