[Bug] GrayToBinary silently produces wrong output when given a color (BGR/BGRA) image

[Jayant-kernel]

Description

The GrayToBinary operator assumes its input is already a single-channel grayscale image, but performs no validation or conversion. When a user connects any BGR or BGRA-output block directly to GrayToBinary, OpenCV's cv2.threshold runs independently on each channel, producing a 3-channel output instead of a true binary grayscale image.

Steps to Reproduce

import cv2, numpy as np
from app.operators.conversions.gray_to_binary import GrayToBinary

# Simulates a BGR image coming from any color operator
img_bgr = np.array([[[50, 150, 200]]], dtype=np.uint8)
result = GrayToBinary({"thresholdValue": 100, "maxValue": 255}).compute(img_bgr)
print(result.shape)   # (1, 1, 3) — NOT a binary grayscale image
print(result)         # [[[  0 255 255]]] — each channel thresholded independently

Expected Behaviour

GrayToBinary should either:

• Raise a ValueError with a clear message if the input is not single-channel (guiding the user to insert GrayImage first), or
• Convert BGR/BGRA to grayscale before thresholding (like ColorToBinary does)

Actual Behaviour

Silently thresholds each colour channel independently, returning a 3-channel array. The block name implies a binary (single-channel) output but the output is a 3-channel image with per-channel thresholding applied — incorrect and misleading.

Context

• ColorToBinary correctly converts to grayscale before thresholding.
• GrayToBinary is intended for images already in grayscale, but the pipeline UI does not prevent connecting a colour-output block to it.
• Related fix: [Bug] GrayToBinary default maxValue=0 causes all-black output #180 (fixed default maxValue), but the colour-input path was not addressed.