LibJpegWrap follow-ups from the review of #2 (Rgb24 codec) #3

These came up in the review of #2 and were left for later:

1. Line endings: src/LibJpegWrap/LibJpegWrap.cpp changed from CRLF to LF in Rgb24 JPEG decode/encode in LibJpegWrap and JpegCodec (for v1.5.0) #2, so its diff is a whole-file rewrite. Decide on a line ending for the file (.gitattributes) and normalise it.
2. Possible double free on the encode error path: in EncodeRgb24, if libjpeg-turbo's jpeg_mem_dest has grown its buffer and an error then longjmps out, free(mem) may free a buffer that empty_mem_output_buffer already freed. The caller's mem is only updated in term_destination. Track the current buffer explicitly, or use a custom destination manager over the caller's buffer.
3. A truncated scan decodes as success: libjpeg treats premature end of data as a warning and fills the rest with gray, so DecodeRgb24 (and the Gray8/I420 decoders) return a full frame. Consider failing when num_warnings > 0, or exposing the warning count.
4. int overflow on huge sizes: MaxJpegSize uses checked and throws, but width * height * 3 in DecodeRgb24 (and FrameHeader lengths elsewhere) can overflow int for very large images. Validate the dimensions before allocating.

Also outside this list: the linux-arm64 and win-x64 LibJpegWrap binaries lack the Rgb24 exports, so the managed calls throw NotSupportedException there until those binaries are rebuilt.

🤖 Generated with Claude Code

https://claude.ai/code/session_014wH5aL3TZhJNvLqa8YTku8