fix: flip spectrogram frequency axis to match freq_bins ordering (#53)

The server data has row 0 = highest frequency, but freq_bins property
calculates assuming row 0 = lowest frequency. This caused frequency
cropping to select the wrong portion of the spectrum (e.g., requesting
200-400 Hz would actually return high frequencies like 2000-2400 Hz).

Now flip the data immediately after loading in from_audio_id() so:
- Row 0 = lowest frequency (matching freq_bins[0])
- Row -1 = highest frequency (matching freq_bins[-1])

This fixes frequency alignment between spectrograms and pitch contours
in visualization contexts.

Note: Visualizations using this API should remove any flipud() calls
they were using to work around this issue.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-authored-by: Claude <noreply@anthropic.com>

Author: jon-myers

idtap/spectrogram.py

@@ -111,6 +111,10 @@ def from_audio_id(cls, audio_id: str, client: Optional['SwaraClient'] = None) ->
         shape = tuple(metadata['shape'])  # [freq_bins, time_frames]
         data = np.frombuffer(decompressed, dtype=np.uint8).reshape(shape)
 
+        # Flip frequency axis so row 0 = lowest frequency (matches freq_bins ordering)
+        # Server data has row 0 = highest frequency, but we want row 0 = lowest
+        data = np.flipud(data)
+
         # Get exact audio duration from recording database
         time_resolution = None
         try: