Large images are not handled consistently

[nicodh]

When sending large images (more than 128 MegaPixel) there are 2 issues:

1. when sent as viewType Image ImageReader runs out of memory when trying to decode the image https://github.com/chatmail/core/blob/main/src/blob.rs#L360 which results in the fallback to viewType File

2. when the image is sent as viewType File core converts it to an image based on the mime type and "converts" it to viewType Image but does NOT run imgreader.decode() and does not rescale the image
   https://github.com/chatmail/core/blob/main/src/blob.rs#L356

See deltachat/deltachat-desktop#5053 and deltachat/deltachat-desktop#6430

The desktop issue has an example image attached

[iequidoo]

1. when sent as viewType Image ImageReader runs out of memory when trying to decode the image https://github.com/chatmail/core/blob/main/src/blob.rs#L360 which results in the fallback to viewType File

I agree that this isn't good. If an image can't be decoded for whatever reason, it can be sent as File, but only if doesn't contain Exif (unfortunately, removing Exif requires recoding currently) and doesn't exceed the byte size limit. We shalln't waste user's traffic.

2. when the image is sent as viewType File core converts it to an image based on the mime type and "converts" it to viewType Image but does NOT run imgreader.decode() and does not rescale the image
   https://github.com/chatmail/core/blob/main/src/blob.rs#L356

If an image is sent as "File" by the user, it doesn't need to be recoded and, correspondingly, decoded. Maybe it's not good that we don't check whether the image is well-formed in this case, but we change its viewtype to Image (optimistically) so that UIs display it as image. Note that we still try to guess the image format from its content or, if that fails, from the file path (extension) (so it can be guessed wrongly).

Not sure we want to fix anything in 2.

EDIT: We have

pub const BALANCED_IMAGE_BYTES: usize = 500_000;

(and also WORSE_IMAGE_BYTES) but it's a "soft" limit for images, i.e. if after recoding an image still doesn't fit into it, we send it anyway. I.e. we recode images only once. Still, if we fail to recode an image, it may be unexpectedly big, we shouldn't send the original if it's hundreds of megabytes, maybe a hard limit of 1 MB makes sense.

EDIT: Wrt 1., we can send an image as Image even if decoding fails so that the behavior is consistent, but it's unlikely that it will be displayed by UIs. Another way to consistent behavior is to always decode ("check") an image before changing the viewtype to Image, but we probably don't want extra CPU work, that's why we only decode for recoding.