Observatory: reduce time with locked mutex

Reduces the time the messages mutex is held by re-arranging some function calls. Also fixes that in clear messages, message_count_ was not reset. !611 (merged) should be rebased on this change since in either case it would benefit as well.

I think in theory, there is still the faint possibility for a data race when clear is pressed and a message arrives, but I think it is highly unlikely:

• in data(), message_count_.load() returns a number > 0
• in clearMessages(), message_count_ is set to 0
• clearMessages() locks the mutex before data() for some very unlucky reason
• clearMessages() clears messages
• data() accesses messages_ with an invalid index