Update calculators for image metrics

[johnwhobbs]

Required Changes

1. Fix HFD Double-Background Subtraction & Centroid Mismatch

• Current Problem: DAOStarFinder(data - median) is used for detection, but _calculate_hfd(data, sources, median) passes the raw, un-subtracted data array and manually subtracts the background in a local slice.
• Update Needed:
  • Calculate bg_subtracted_data = data - median globally right after computing sigma_clipped_stats.
  • Pass bg_subtracted_data directly into DAOStarFinder.
  • Update the signature of _calculate_hfd to accept bg_subtracted_data and remove the background_median argument entirely. Slices inside _calculate_hfd must come directly from this pre-subtracted array.

2. Upgrade Color Channel Reduction to True Luminance (CIE Weights)

• Current Problem: data = data.mean(axis=channel_axis) weighs Red, Green, and Blue channels uniformly, which degrades star profiles on color frames.
• Update Needed: Replace uniform averaging with a standard CIE luminance dot product ($Y = 0.299R + 0.587G + 0.114B$) when processing a 3-channel array. If the channel axis is at index 0, properly reorient or handle the axis shift before applying the matrix weights.

3. Update Star Roundness to Track Maximum Asymmetry

• Current Problem: The code concatenates roundness1 and roundness2 and takes the uniform np.mean(), which dilutes and hides directional tracking errors (like a sudden wind gust or periodic error along one axis).
• Update Needed: Change the aggregation logic from np.mean() to np.max(). The final star_roundness metric must reflect the maximum absolute asymmetry detected across the profiles to ensure strict quality control on tracking.

4. General Cleanup & Performance Guardrails

• Ensure that clipping negative pixels to zero (cutout[cutout < 0] = 0) happens safely on an explicit copy of the array slice inside _calculate_hfd to prevent modifying or raising warnings on view slices.
• Ensure type casting to standard Python floats/ints remains intact before writing to the database dictionary wrapper.

[johnwhobbs]

✅ Pull request created with the requested changes.