If file has been modified since the cache, regenerate a new cache
image.
We don't want to check the file modified time too often though,
(e.g., we don't want to check for every point render in a 100k
point file), so use a hardcoded 30 second minimum time between
consecutive file modified checks.
This means that file modifications occuring more often than
every 30 seconds won't be picked up till 30 seconds has elapsed
since the last modification. But at the same time it means that
if the render takes < 30 seconds we'll only check each svg
at most once (and if a render takes > 30 seconds, adding a few
more milliseconds won't hurt!).
Fixes#13565
QgsSvgCache::svgAsPicture was rendering an implicitly shared copy when
the picture had already been cached. It turns out that rendering an
implicitly shared QPicture copy isn't thread safe, and rendering shared
copies simulataneously across different threads leads quickly to a crash.
Accordingly we always detach the QPicture objects returned by
svgAsPicture, so that the returned QPicture is safe to use across threads.
Also add unit tests for this, and a similar unit test to verify that
rendering of QImage based cached copies does *not* suffer the same
issue.
Fixes#17089, #17077
