Fix missing image_size parameter in eval.py

[aug6th]

The eval() method was calling _eval() with only 4 parameters (coordinate, boxes_type, boxes_size, boxes_coordinate) on line 31, but _eval() requires 5 parameters including image_size (line 35-41).

This caused a TypeError when the eval() method was called:
TypeError: _eval() missing 1 required positional argument: 'image_size'

The image_size parameter is essential because _eval() uses it on line 67 to scale normalized coordinates (values between 0 and 1) to pixel coordinates.

This fix adds:

1. Line 30: Extract image_size from the PIL Image object
2. Line 32-34: Pass image_size as the 5th parameter to _eval()

This brings eval.py into consistency with all other evaluation scripts (qwen25_vllm_osworld_g_jedi.py, screenspot_v2, screenspot_pro, aguvis, operator) which all correctly pass image_size to _eval().

Note: This bug only affected the eval() method used for standalone testing (line 90-92). The _eval() method itself was always correct and is used properly by all production evaluation scripts.