Only attempt to re-attach if the management process stopped. On
errors such as overrun or log abandoned, the VSL API considers
the client as still attached, and VSM_Attach() throws an assertion.

If the process has changed or restarted, re-initialize the shared
memory attachment and re-acquire the log. A new reader metric
mgt_restart is incremented when the status change is detected.

We run the VSM_Status() check when end-of-log is reached, or when
a flush has been signaled, to minimize the hit from the fstat()
calls in its implementation.

None of this is done when reading from a log file.

Never, ever, copy the docs.

By re-ordering #include vdef.h.

The varnish package (from packagecloud.io) makes the VSM directories
non-world-readable, but readable to the group varnish.

So the RPM no longer creates the group trackrdrd, but adds the
trackrdrd to group varnish.

No git tag yet, testing automated packaging with version number.

Into separate files.

This requires a user who can write in /var/log. So we now create
user/group trackrdrd at pre-install, and the default config runs
with that user for the child process.