Frame on Frame RAM Increase Using PyBindings for 3.1.0 Converting EXR To Uncompressed PNG

I'm using the Python Bindings for 3.1.0 with Python 3.10.6 on Windows and getting increasing RAM usage with the script below which converts exrs to pngs - about 1Gb for every frame processed. For context, the original .exr is circa 20k x 2k so is pretty large. This doesn't seem to be a problem for oiiotool.exe (3.1.0) which does the same process perfectly with no increase in RAM usage frame by frame. It's also not a problem if I convert from .exr to .tiff so it does seem to be specific to png. Obvious answer is to switch from png to tiff which I will do but felt I should at least make a record of this here.

```
import OpenImageIO as oiio

def convert_exr_to_png(exr_file, png_file):
# Open the EXR file
input_image = oiio.ImageInput.open(exr_file)
if not input_image:
print("Error opening EXR file:", oiio.geterror())
return

# Read the image data
spec = input_image.spec()
pixels = input_image.read_image()
input_image.close()

# Create the output PNG file
output_image = oiio.ImageOutput.create(png_file)
if not output_image:
print("Error creating PNG file:", oiio.geterror())
return

# Set the compression level to 0 (no compression)
spec.attribute("png:compressionLevel", 0)
spec.attribute("png:filter", 8)
#spec.attribute("oiio:ColorSpace", "sRGB")

# Open the output PNG file
output_image.open(png_file, spec)

# Write the image data to the PNG file
output_image.write_image(pixels)

# Close the output image
output_image.close()
pixels = None
print(f"Successfully converted {exr_file} to {png_file} with no compression.")

for x in range(1, 2701):
frame = str(x).zfill(4)
exr = f"//flabby/jobs/client_62185/build/virtual_set/comps/openimage/stitchhalfres/stitch_v021.{frame}.exr"
png = f"//flabby/jobs/client_62185/build/virtual_set/comps/openimage/stitchdebug/stitchcspace_v021.{frame}.png"
convert_exr_to_png(exr, png)
print("Processed Frame: ", x)
```

Start with the supplied Python conversion script and reproduce repeated EXR-to-uncompressed-PNG conversions on Windows, comparing memory use with EXR-to-TIFF and oiiotool.exe. Investigate the Python binding and PNG path to determine why RAM grows between frames; done means identifying the cause and confirming stable memory across repeated conversions.

Written by the indexing model from the issue text.