fix: avoid erroring out on corrupt image data

[joshtrichards]

Fixes #2257
Fixes #2235

As an aside, we should consider switching from using getimagesizefromstring() to getimagesize() if we can -- it consumes far less memory in my testing (presumably since the entire image isn't loaded into a string in memory).

P.S. Technically this also addresses the logging side of #2259, but it's unclear whether that's arising from a memory limit or what. I can only get that error if I feed getimagesizefromstring() /dev/null. :-) However this PR does add logging of the path so it'll likely help work out that issue as well.

[joshtrichards]

/backport to stable29

[joshtrichards]

/backport to stable28

[artonge]

As an aside, we should consider switching from using getimagesizefromstring() to getimagesize() if we can

I think to recall that getimagesize() would not work well with external storage. But maybe I am wrong.