We were implicitly relying on mongo to return the list of members in
their _id order, which is absurd and I have no idea how we didn't catch
it previously.

We could do a little list comprehension to pull out the member with the
same _id,  but it seemed easier to just sort both lists and retain the
original intent.

I believe the loops will be easier to debug in the long run.
Essentially, we have a shallow compare loop in check_config_subset
and then a deep compare loop in is_member_subset.

Both functions allow extra information in the config that comes back
from Mongo, since Mongo tracks things like votes: slaveDelay: etc that
may not be reflected in the configuration you pass in.

Do a bit more error checking before asking for the replica set.

The new mongo play wants to be able to run this against an unconfigured
Mongo to find out "Do you even have a replica set?" which this can now
answer (the try/except portion).

In the case where mongo was started without a replSet configured (or
running getCmdLineOpts fails for an OperationFailure reason), we
return a dictionary where status is not included to allow tasks to
detect the failure.  Unfortunately, Ansible doesn't provide an easy way
to log back why you didn't get a replica set doc.

Documentation improvements for this module

Discusses the return values and some examples of checking them.

Simplify get_replset

If the find_one for a replset fails, just return None.
We don't need to worry about getting back a dictionary which contains a
'config' key (presumably this is from an earlier iteration of the code
which called getReplSetStatus which does act like that).

Rather than raising the exception, use the friendly error here.

I assume the raise/module.fail_json double header was leftover cruft
from earlier development.

Make the module work without rs being configured

Initially, this module always tried to find and connect to a primary,
however, during replica set initialization, there is no primary so we
need to fall back to just connecting to the machine specified.

This is detected by catching the replSetGetStatus failure and falling
back to the bare get_client method which refactors the connection.

Now you can provide a (slightly simpler) cluster config and it'll
idempotently make it so.