feat: Add themed FFT waveform visualization to detail panel

Backend:
- Install ffmpeg + numpy in Docker image
- New waveform.py: FFT-based SVG generator with 64 frequency bands
- GET /api/clips/{id}/waveform — generates on demand, caches to /data/waveforms/
- DELETE /api/clips/{id} now cascade-removes waveform SVG
- Warm amber-to-orange HSL gradient (25° to 35° hue)
- Mirrored bars: top full opacity, bottom 60% opacity

Frontend:
- DetailPanel shows waveform image beneath album art (80px height)
- Only shown for clips with status='ready'

Storage: /data/waveforms/ persisted via existing ./data volume mount

backend/src/main.py

@@ -12,6 +12,7 @@ load_dotenv(Path(__file__).parent.parent / ".env")
 from models import GenerateRequest, GenerateResponse, ClipList, RenameRequest, LYRIA_MODELS, DEFAULT_MODEL
 from database import init_db, create_clip, get_clips, get_clip, delete_clip, update_clip_status, rename_clip, set_artwork, remove_artwork
 from lyria_client import generate_clip, save_clip_file
+from waveform import generate_waveform_svg, get_waveform_path
 
 app = FastAPI(title="Lyria Studio", version="0.1.0")
 
@@ -124,10 +125,33 @@ async def remove_clip(clip_id: str):
         artwork_path = ARTWORK_DIR / clip["artwork_filename"]
         if artwork_path.exists():
             artwork_path.unlink()
+    # Also delete waveform if it exists
+    waveform_path = get_waveform_path(clip_id)
+    if waveform_path and waveform_path.exists():
+        waveform_path.unlink()
     await delete_clip(clip_id)
     return {"deleted": True}
 
 
+@app.get("/api/clips/{clip_id}/waveform")
+async def get_waveform(clip_id: str):
+    clip = await get_clip(clip_id)
+    if not clip:
+        raise HTTPException(status_code=404, detail="Clip not found")
+    # Check if waveform already exists
+    svg_path = get_waveform_path(clip_id)
+    if svg_path and svg_path.exists():
+        return FileResponse(svg_path, media_type="image/svg+xml")
+    # Generate on demand
+    try:
+        svg_path_str = generate_waveform_svg(clip_id, clip["filename"])
+        return FileResponse(svg_path_str, media_type="image/svg+xml")
+    except FileNotFoundError:
+        raise HTTPException(status_code=404, detail="Audio file not found")
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Waveform generation failed: {str(e)}")
+
+
 @app.get("/api/clips/{clip_id}/artwork")
 async def get_artwork(clip_id: str):
     clip = await get_clip(clip_id)

backend/src/waveform.py

@@ -0,0 +1,154 @@
+import subprocess
+import numpy as np
+from pathlib import Path
+
+CLIPS_DIR = Path(__file__).parent.parent / "data" / "clips"
+WAVEFORM_DIR = Path(__file__).parent.parent / "data" / "waveforms"
+WAVEFORM_DIR.mkdir(parents=True, exist_ok=True)
+
+
+def generate_waveform_svg(clip_id: str, filename: str) -> str:
+    """Generate a themed FFT waveform SVG for an audio clip.
+    
+    Returns the path to the generated SVG file.
+    """
+    audio_path = CLIPS_DIR / filename
+    svg_path = WAVEFORM_DIR / f"{clip_id}.svg"
+    
+    if not audio_path.exists():
+        raise FileNotFoundError(f"Audio file not found: {audio_path}")
+    
+    # Decode audio to raw PCM using ffmpeg
+    # Output: 16-bit signed little-endian, mono, 22050 Hz
+    cmd = [
+        "ffmpeg", "-i", str(audio_path),
+        "-f", "s16le", "-acodec", "pcm_s16le",
+        "-ac", "1", "-ar", "22050",
+        "-"
+    ]
+    
+    result = subprocess.run(cmd, capture_output=True)
+    if result.returncode != 0:
+        raise RuntimeError(f"ffmpeg failed: {result.stderr.decode()[:200]}")
+    
+    # Convert raw bytes to numpy array
+    audio_data = np.frombuffer(result.stdout, dtype=np.int16)
+    
+    # Normalize to float [-1, 1]
+    audio_data = audio_data.astype(np.float32) / 32768.0
+    
+    # Parameters
+    num_bands = 64
+    svg_width = 800
+    svg_height = 200
+    bar_gap = 2
+    bar_width = (svg_width - (num_bands - 1) * bar_gap) / num_bands
+    max_bar_height = svg_height * 0.9  # Leave some margin
+    
+    # Divide audio into segments and compute FFT for each
+    samples_per_band = len(audio_data) // num_bands
+    
+    band_heights = []
+    for i in range(num_bands):
+        start = i * samples_per_band
+        end = start + samples_per_band
+        segment = audio_data[start:end]
+        
+        if len(segment) < 2:
+            band_heights.append(0)
+            continue
+        
+        # Apply Hann window
+        window = np.hanning(len(segment))
+        windowed = segment * window
+        
+        # FFT
+        fft = np.fft.rfft(windowed)
+        magnitude = np.abs(fft)
+        
+        # Use the average magnitude in the lower-mid frequencies (most audible)
+        # Focus on ~100Hz - 4kHz range
+        freqs = np.fft.rfftfreq(len(segment), d=1.0/22050)
+        mask = (freqs >= 100) & (freqs <= 4000)
+        if np.any(mask):
+            avg_mag = np.mean(magnitude[mask])
+        else:
+            avg_mag = np.mean(magnitude[:len(magnitude)//4])  # fallback: first quarter
+        
+        # Convert to dB-like scale and normalize
+        # Add small offset to avoid log(0)
+        db = 20 * np.log10(avg_mag + 1e-10)
+        # Map roughly -60dB to 0dB range to 0-1
+        normalized = (db + 60) / 60
+        normalized = np.clip(normalized, 0, 1)
+        
+        band_heights.append(normalized)
+    
+    # Smooth the heights slightly for visual appeal
+    smoothed = []
+    for i, h in enumerate(band_heights):
+        neighbors = band_heights[max(0, i-1):min(len(band_heights), i+2)]
+        smoothed.append(np.mean(neighbors))
+    band_heights = smoothed
+    
+    # Build SVG
+    # Color gradient: warm amber/orange tones (lyria-accent themed)
+    # Using HSL interpolation from deep amber to bright orange
+    def band_color(index: int, total: int) -> str:
+        """Generate color for a band — warm amber to orange gradient."""
+        t = index / (total - 1) if total > 1 else 0
+        # HSL: hue 25° (amber) to 35° (orange), saturation 80-90%, lightness 45-65%
+        hue = 25 + t * 10
+        sat = 85
+        light = 45 + t * 20
+        return f"hsl({hue:.1f}, {sat}%, {light:.1f}%)"
+    
+    # SVG content
+    svg_parts = [
+        f'<svg xmlns="http://www.w3.org/2000/svg" width="{svg_width}" height="{svg_height}" viewBox="0 0 {svg_width} {svg_height}" preserveAspectRatio="none">',
+        '<defs>',
+        '  <linearGradient id="barGrad" x1="0%" y1="0%" x2="0%" y2="100%">',
+        '    <stop offset="0%" stop-color="#f59e0b" />',   # amber-500
+        '    <stop offset="100%" stop-color="#ea580c" />',  # orange-600
+        '  </linearGradient>',
+        '</defs>',
+        f'<rect width="{svg_width}" height="{svg_height}" fill="transparent" />',
+    ]
+    
+    center_y = svg_height / 2
+    
+    for i, height in enumerate(band_heights):
+        x = i * (bar_width + bar_gap)
+        bar_h = height * max_bar_height / 2  # Half goes up, half goes down
+        
+        # Color based on position in spectrum
+        color = band_color(i, num_bands)
+        
+        # Top bar (above center)
+        svg_parts.append(
+            f'  <rect x="{x:.2f}" y="{center_y - bar_h:.2f}" '
+            f'width="{bar_width:.2f}" height="{bar_h:.2f}" '
+            f'fill="{color}" rx="{bar_width/4:.2f}" />'
+        )
+        
+        # Bottom bar (mirror, below center)
+        svg_parts.append(
+            f'  <rect x="{x:.2f}" y="{center_y:.2f}" '
+            f'width="{bar_width:.2f}" height="{bar_h:.2f}" '
+            f'fill="{color}" rx="{bar_width/4:.2f}" opacity="0.6" />'
+        )
+    
+    svg_parts.append('</svg>')
+    
+    svg_content = '\n'.join(svg_parts)
+    svg_path.write_text(svg_content, encoding='utf-8')
+    
+    return str(svg_path)
+
+
+def get_waveform_path(clip_id: str) -> Path | None:
+    """Get path to existing waveform SVG, or None if not generated."""
+    svg_path = WAVEFORM_DIR / f"{clip_id}.svg"
+    if svg_path.exists():
+        return svg_path
+    return None