In some scenarios, a crashing process can resulted in a corrupted cache file.
This strengthens the code by deleting the shared memory files on startup to guarantee a clean start.
This should already happen on other OSes.
Test using a shared memory file with am->global_size set incorrectly. When agent is down, put the files in the agent/logs directory
Agent should always start cleanly and not crash on a request
Agent crashes with a c0005 access violation in am_shm_lock+037