gifsave: ensure return code is propagated

[lovell]

Whilst investigating lovell/sharp#4257 I discovered that vips_sink_disc appears to always return zero during vips_foreign_save_cgif_build even when vips_foreign_save_cgif_sink_disc returns a non-zero return code. It feels like a bit of a race condition, and I think this is because we need to have the entire input in memory before the CPU-intensive quantisation takes place.

This PR adds a new aborted property to track any non-zero return code, which then allows for things like an eval callback. The call to vips_image_eval occurs for each frame just after the image is remapped to the new palette as this is the most CPU-intensive section. I'd be happy to move the vips_image_eval call to a different location if required, or perhaps it might be better to add multiple calls given the CPU-intensive nature of quantisation as a whole.

[jcupitt]

I did a little digging. sink_disc picks up error returns here:

https://github.com/libvips/libvips/blob/master/libvips/iofuncs/sinkdisc.c#L174

And handles them here:

https://github.com/libvips/libvips/blob/master/libvips/iofuncs/sinkdisc.c#L264

But it does this lazily, since the write is actually in a background thread. It can't detect a write error until the background thread that was doing the write terminates.

The wait for the exit of the final write happens here:

https://github.com/libvips/libvips/blob/master/libvips/iofuncs/sinkdisc.c#L532

... but we've already set the return code on line 518, so a failure in the final write is never returned.

A simple fix might be to change the end of vips_sink_disc() to be:

    /* Just before allocate signalled stop, it set write.buf writing. We
     * need to wait for this write to finish.
     *
     * We can't just free the buffers (which will wait for the bg threads
     * to finish), since the bg thread might see the kill before it gets a
     * chance to write.
     *
     * If the pool exited with an error, write.buf might not have been
     * started (if the allocate failed), and in any case, we don't care if
     * the final write went through or not.
     */
    if (!result)
        vips_semaphore_down(&write.buf->done);

    vips_image_posteval(im);

    /* The final write might have failed, pick up any error code.
     */
    result |= write.buf->write_errno;

    write_free(&write);

    vips_image_minimise_all(im);

    return result;
}

ie. OR in any code from the final thread. Though maybe we should also set an error message? Flush does this:

        /* Previous write succeeded?
         */ 
        if (write->buf_back->write_errno) {
            vips_error_system(write->buf_back->write_errno,
                "wbuffer_write", "%s", _("write failed"));
            return -1;
        }   

So we should do the same for the final write, or maybe remove the message from flush.

[jcupitt]

I think you can write vips_image_eval(cgif->in, n_pels); here