When a storage child actor gets the xpcom-shutdown notification, it releases
the actor singleton which asynchronously sends __delete__ to the parent and
destroys the child actor immediatelly. However, at the same time parent can be
sending a message to the child which results in a routing error since the child
actor doesn't exist anymore. The routing error causes a crash.
This patch prevents the race by doing the teardown in two steps. We send a
message to the parent DeleteMe and parent then sends __delete__ to the child.
- stop inheriting StorageDBBridge in StorageDBThread and StorageDBChild
- move StorageDBThread and StorageDBChild initialization out of LocalStorageCache
- use IPC even for the intra-process communication in main process
- rationalize a bit storage observer code
- make StorageDBParent to always be created and destroyed on the background thread
