sync issue boolean and gui->reset

In the lighttable view, the thread was 'worker' and 'thumbs_update*'.

It looks to me like the darktable thread and thumbs_update can run in parallel. When I run dt 5.4.1 in gdb and set an awatch on darktable.gui->reset I see stuff like this

Thread 1 "darktable" hit Hardware access (read/write) watchpoint 1: darktable.gui->reset

Value = 0

Thread 1 "darktable" hit Hardware access (read/write) watchpoint 1: darktable.gui->reset

Value = 0
0x00007ffff7af91ff in dt_view_image_get_surface (imgid=304, width=98, height=80, surface=surface@entry=0x7fffffff6fd0, quality=quality@entry=0) at /home/david/devel/darktable/darktable/src/views/view.c:727
727       *surface = NULL;
[Switching to Thread 0x7fffe59fe6c0 (LWP 777633)]

Thread 20 "thumbs_update" hit Hardware access (read/write) watchpoint 1: darktable.gui->reset

Old value = 0
New value = 1

Thread 20 "thumbs_update" hit Hardware access (read/write) watchpoint 1: darktable.gui->reset

Value = 1
dt_iop_reload_defaults (module=0x7fff9800afd0) at /home/david/devel/darktable/darktable/src/develop/imageop.c:1299
1299      if(module->reload_defaults)
[Switching to Thread 0x7fffef1a4f40 (LWP 777612)]

Thread 1 "darktable" hit Hardware access (read/write) watchpoint 1: darktable.gui->reset

Value = 1

Thread 1 "darktable" hit Hardware access (read/write) watchpoint 1: darktable.gui->reset

Value = 1

Without synchronised access to darktable.gui->reset those two threads (and maybe others?) are bound to get tangled up, aren't they?

Originally posted by @dhoulder in #21043

Start by examining darktable.gui-reset accesses around dt_view_image_get_surface in src/views/view.c and dt_iop_reload_defaults in src/develop/imageop.c. Reproduce the reported watchpoint activity under gdb, identify all participating threads and accesses, and consider the issue done when the shared state is accessed safely without the reported race.