Watchdog interferes with cron semaphore deletion in drupal_cron_cleanup()

While debugging some problems with cron, I realized that the watchdog() call in drupal_cron_cleanup() can cause problems in some cases. The drupal_cron_cleanup() function is called at the end of a cron run, including in situations where cron failed due to a PHP timeout error or other serious problem. In that case, things might be in a very weird state. The watchdog() call might therefore lead to fatal PHP errors of its own. (The particular case where I saw this occur is if you have Syslog and Views UI modules enabled, but I'm sure there are others.)

Since the most important thing that drupal_cron_cleanup() does is delete the cron semaphore, my solution was just to move that up to the top of the function. That way if things fail catastrophically later on, at least the semaphore is deleted, and cron is not locked up for the next hour. (If you look through the forums, they're full of people who are getting cron locked up for an hour and need to delete the semaphore by hand in the database, etc. I don't think this will solve all of those problems, but it should help in some cases -- and certainly it doesn't hurt!)

Patch is attached. I also modified the message that watchdog() prints in this case, since it simply isn't true that the only way for the semaphore to be locked is the if the PHP script execution time was exceeded. Certain other types of errors (e.g., calls to nonexistent functions) during the cron run can trigger this too.