Updates can deadlock if state is being closed concurrently

[adamgundry]

It isn't terribly well specified what is supposed to happen if closeAcidState is called concurrently with an update transaction. The typical outcomes are that the update will either succeed or throw an error "Access failure: Core closed." While it would be nicer if this was a custom exception type rather than ErrorCall, this is fine.

However, sometimes a deadlock can result, with "FileLog has been closed" printed on the console (from the acid-state background writer thread, I think) and the update never returning. This seems wrong to me: I think the update should either succeed or throw an exception.

This isn't a big problem in typical server applications, because usually closeAcidState will be called right before program shutdown. But it might be a problem if an application wanted to close the state and keep working. Moreover, it makes it a pain to test for other race conditions involving closeAcidState and related operations.

I've put together a program demonstrating this.