Auto blue red gains

[mimanerhn]

Changes in the analyze_image.py script

1. New CLI flag

• --gain_search <N>
  An integer option (default 0) that, when used with --bayer, triggers the single‐shot quantile‐fitting auto-gain routine if N > 0; 0 (the default) preserves the original single-pass behavior.

2. Auto-gain logic (auto_gain_locally)

1. Initialization

   • Use provided --rg/--bg if positive, otherwise default to (1.0, 1.0).
   • Load raw Bayer data via load_image(fp, verbose=False, bayer=True, return_float=False, …) and extract the single-channel array.

2. Quantile fitting
   We pick a handful of brightness cutoffs (roughly the 65th–95th percentiles) and compute one red gain and one blue gain that best match those levels to green (no while or loops needed).

   • Why percentiles?
     They focus on mid-range tones and avoid extreme darks or brights, which can be noisy or clipped.

   • How it works

     1. Measure the chosen brightness cutoffs for red, green, and blue.
     2. Compute a single multiplier for red and for blue that lines those cutoffs up with green.
     3. Apply those gains for a balanced color fit in one pass.

3. Gain application

   • Clamp the fitted gains to [0.5, 2.5].
   • Log the result:

     [AutoGain] fitted gains → rg=1.345, bg=1.078
     

4. Final pass

   • Invoke analyze_image(...) once more with the new gains (and gain_search=0) to produce the standard plots and outputs.

3. Entry-point dispatch (analyze_image)

• If --bayer and --gain_search N>0, dispatch to auto_gain_locally(...).
• Otherwise, run normal analysis with the provided or default gains.