Add test suite for Lyria Studio backend

- conftest.py: isolated async test fixtures with temp DB and directories
- test_waveform.py: waveform generation, downsampling, caching, SVG output
- test_api.py: full API endpoint tests (health, models, clips CRUD,
  waveform endpoint, admin cache clearing)
- test_analysis.py: BPM detection, key detection, clip analysis
- test_models.py: request validation, model registry, prompt building

- Fixed off-by-one in waveform.py downsampling (linspace upper bound)
- All 40 tests passing

backend/tests/test_models.py

@@ -0,0 +1,90 @@
+"""Tests for Pydantic models."""
+
+import sys
+from pathlib import Path
+
+import pytest
+
+sys.path.insert(0, str(Path(__file__).resolve().parent.parent / "src"))
+
+from models import GenerateRequest, GenerateResponse, ClipList, LYRIA_MODELS, DEFAULT_MODEL
+
+
+class TestGenerateRequest:
+    """Request validation and prompt building."""
+
+    def test_default_model_and_duration(self):
+        req = GenerateRequest(prompt="gabber kick")
+        assert req.model == DEFAULT_MODEL
+        dur = req.get_effective_duration()
+        assert dur == 30  # clip preview default
+
+    def test_pro_model_with_custom_duration(self):
+        req = GenerateRequest(prompt="180 BPM hardcore", model="lyria-3-pro-preview", duration=120)
+        assert req.model == "lyria-3-pro-preview"
+        dur = req.get_effective_duration()
+        assert dur == 120
+
+    def test_pro_model_duration_clamped_to_max(self):
+        req = GenerateRequest(prompt="long track", model="lyria-3-pro-preview", duration=999)
+        dur = req.get_effective_duration()
+        assert dur == 184  # max for pro model
+
+    def test_clip_model_duration_clamped_to_max(self):
+        req = GenerateRequest(prompt="short clip", model="lyria-3-clip-preview", duration=60)
+        dur = req.get_effective_duration()
+        assert dur == 30  # max for clip preview
+
+    def test_invalid_model_raises(self):
+        with pytest.raises(ValueError, match="Unknown model"):
+            GenerateRequest(prompt="test", model="nonexistent-model")
+
+    def test_simple_prompt(self):
+        req = GenerateRequest(prompt="gabber at 200 BPM")
+        assert req.build_prompt() == "gabber at 200 BPM"
+
+    def test_structured_prompt(self):
+        req = GenerateRequest(
+            prompt="ignored",
+            prompt_structure={
+                "style": "gabber",
+                "tempo": "180 BPM",
+                "mood": "aggressive",
+            }
+        )
+        built = req.build_prompt()
+        assert "**Style:** gabber" in built
+        assert "**Tempo:** 180 BPM" in built
+        assert "**Mood:** aggressive" in built
+
+    def test_model_whitespace_trimmed(self):
+        req = GenerateRequest(prompt="test", model=" lyria-3-clip-preview ")
+        assert req.model == "lyria-3-clip-preview"
+
+
+class TestModelConstants:
+    """Validate the model registry."""
+
+    def test_models_have_required_keys(self):
+        for model_id, info in LYRIA_MODELS.items():
+            assert "name" in info, f"{model_id} missing 'name'"
+            assert "max_duration" in info, f"{model_id} missing 'max_duration'"
+            assert "default_duration" in info, f"{model_id} missing 'default_duration'"
+            assert "description" in info, f"{model_id} missing 'description'"
+
+    def test_default_model_exists_in_registry(self):
+        assert DEFAULT_MODEL in LYRIA_MODELS
+
+    def test_max_duration_greater_than_default(self):
+        for model_id, info in LYRIA_MODELS.items():
+            assert info["max_duration"] >= info["default_duration"], \
+                f"{model_id}: max_duration < default_duration"
+
+
+class TestGenerateResponse:
+    """Response model."""
+
+    def test_create_response(self):
+        resp = GenerateResponse(id="test-123", status="generating", message="Clip generation started.")
+        assert resp.id == "test-123"
+        assert resp.status == "generating"