drive.
Change frontend reporting for the current stats from bytes to Mbytes and
from bytes/s to Mbytes/s.

If the "cta af ls" command is executed and the listing
encounters duplicate tape copy numbers then the listing
continues with "workaround" number displayed in place
of the duplicates and inow with this commit a log
message is alseo written to

    /var/log/cta/cta-frontent.log

An example log:

2017-07-12T18:04:40.691236+02:00 itdssbuild01 cta-frontend: LVL="Warn" PID="9423" TID="9436" MSG="Found a duplicate tape copy number when listing archive files" archiveFileID="5" duplicateCopyNb="1" workaroundCopyNb="2" vid="V41001" fSeq="5" blockId="41"

recall tape transfer session.

TaskWatchDog pereodically update Object store information with
current values for the tape session stats.

If the drive is set to force down mode, we stop finding new jobs and let the current session drain.

The typo made Vlado's eyes bleed.

This will help diagnose failure to honor maxDrivesAllowed.

The queue is now fully event driven: a write to object store will start as soon as the previous one completed (if any).
The notion of timeout and max number of elements before flush has been removed. This harmonises the mechanism with the
one used in MemArchiveQueue::sharedAddToArchiveQueue() (see a591129c).

This constraint prevents a tape server from writing the same file
more than once to the same tape.  We should actually allow this
to happen so that we can tolerate the pathological case of a
cta-taped daemon crashing immediately after inserting a row into
the TAPE_FILE table and before updating the object store.

Implemented GarbageCollector::reinjectOwnedObject() and called it on garbage collector shutdown.
This fixes the issue of non-empty garbage collectors exiting.

Simplifies code and avoids unnecessary string copy construction

We now only do one queue update at a time (per process). While one batch is being pushed to the objet store, the
following one is being built up, and written as soon as the previous write completes. This can be see in the timings from
the logs of the unit test OStoreSchedulerDatabaseTestVFS/SchedulerDatabaseTest.createManyArchiveJobs/0 here:
[...] MSG="In MemArchiveQueue::sharedAddToArchiveQueue(): added batch of jobs to the queue." objectQueue="archiveQueue-[...]-68" sizeBefore="0" sizeAfter="1" addedJobs="1" waitTime="0.002403" enqueueTime="0.036594"
[...] MSG="In MemArchiveQueue::sharedAddToArchiveQueue(): added batch of jobs to the queue." objectQueue="archiveQueue-[...]-68" sizeBefore="1" sizeAfter="100" addedJobs="99" waitTime="0.036547" enqueueTime="0.001629"
[...] MSG="In MemArchiveQueue::sharedAddToArchiveQueue(): added batch of jobs to the queue." objectQueue="archiveQueue-[...]-174" sizeBefore="0" sizeAfter="1" addedJobs="1" waitTime="0.000058" enqueueTime="0.061447"
[...] MSG="In MemArchiveQueue::sharedAddToArchiveQueue(): added batch of jobs to the queue." objectQueue="archiveQueue-[...]-174" sizeBefore="1" sizeAfter="198" addedJobs="197" waitTime="0.061434" enqueueTime="0.001431"
[...] MSG="In MemArchiveQueue::sharedAddToArchiveQueue(): added batch of jobs to the queue." objectQueue="archiveQueue-[...]-174" sizeBefore="198" sizeAfter="200" addedJobs="2" waitTime="0.000840" enqueueTime="0.037259"

twice and outputs EOS workflow proc content to list various type of
errors.

It appeared in production that the update latency could be steadily higher than this delay.
This led to an ever growing perceived latency by the client, as the batches to be queued accumulated
faster than they were drained.

This is an interim solution to a more adaptive one which should make no asumption.

Handling of multi-block responses and streaming is not yet implemented.

Renames private member variables. Changes to some constructor arguments.

First character of XrdSsi metadata is overwritten with zero, see
https://github.com/xrootd/xrootd/issues/537
This workaround adds a dummy character to the front of the metadata on
the server side and removes it on the client side.