Also fixed potential crash within a race condition (os.sleep => time.sleep).

Also removed outdated mammouth instruction.

- the lock is now properly set in the new compiledir when set_compiledir is
  called after an initial call to get_lock()
- timeout is now 2 mins instead of 1 (safer in case compilation takes a while)
- the loop trying to acquire the lock does not raise an exception anymore when
  something weird happens, because who knows what might happen with parallel
  stuff ;)

(either directly or indirectly) to Module is deprecated.  Just use
Variable."

- The lock should now be properly released even after abnormal termination
- Verbosity of the lock has been reduced while waiting to acquire a lock
- Fixed bug where a job would not be able to acquire a lock when the timeout
  expires