[tanjiroK-coder]

TiffDecoder::set_limits reads the image dimensions from self.info, but that field is only populated once prepare_image has run. In the order the ImageDecoder protocol documents, and the order ImageReaderOptions::into_decoder uses, set_limits runs first, so it checked a placeholder of 0x0 and strict max_image_width/max_image_height limits were silently accepted for any TIFF; ImageReader::decode is unaffected because it re-checks the layout, but a decoder handed to DynamicImage::from_decoder decodes at full size. This is a regression from the decoder refactor in e6a182c, which replaced the previous self.dimensions() call with the dummy and adjusted the test to call prepare_image before set_limits. AvifDecoder::set_limits has the same gap in a simpler form: it stores the limits without calling check_support or check_dimensions. Both now look up the real layout the way the trait default and the other decoders do. The prepare_image workaround in tests/limits.rs is removed, so the existing tiff and avif assertions fail on main and pass with this change (the avif test only runs with avif-native, which I built locally via the internal dav1d build).