Fail accumulo-service stop/kill when no service name is given

[DomGarguilo]

Running accumulo-service manager stop without --all or an instance name silently does nothing. The script:

• prints Stopping service process: with a blank name,
• looks for the pid file accumulo-.pid, which doesn't exist,
• exits 0.

The manager keeps running even though the command reported success. A later start then says manager_default_1 already running. kill has the same problem, with no output at all.

This PR adds a check in function main() before stop/kill are dispatched. When all of these are true:

1. no service name was given
2. --all was not passed
3. the script is not in cluster mode (ACCUMULO_CLUSTER_ARG unset, so calls from accumulo-cluster are unaffected)

it exits 1 with this message, followed by the usage text:

Invalid arguments: 'stop' requires --all or a service (e.g. manager_default_1; see 'manager list')

Nothing is stopped, and the user is told what to pass instead.

Also:

• The usage line changes from stop [--all | []] to stop [--all | ] (same for kill), so the name no longer looks optional.
• invalid_args now prints its "Invalid arguments" line to stderr, matching the usage text it already sent there.

[ddanielr]

Is there a reason not to add this arg check in the logic for the stop_service and kill_service functions?

[dlmarion]

I wonder if the absence of a more specific parameter should imply 'all'. I'm thinking about other scripts and how they behave for consistency. For example, if you run accumulo-cluster stop, then entire cluster is stopped. The user has to supply arguments for a more targeted action. Maybe we should do the same here?

[ddanielr]

Originally this ability was added to target names of processes that had been left over from previous cluster actions.
For example: tserver_default_4 would be running but not targetable by stop/kill since the cluster.yaml had been updated to only run 2 tserver processes.

I'm hesitating about adding an easy way to stop all services without forcing the user to be intentional.
accumulo-cluster stop makes sense because cluster implies that multiple nodes/services are being targeted.

[DomGarguilo]

Yea it seems safer force the user to be more explicit when the operation is destructive.

[ddanielr]

The behavior change is beneficial. I'd switch this PR to the 2.1 branch though.
There was a good bit of work done to make sure the control scripts matched between 2.1 and main.

[DomGarguilo]

I moved things into stop_service and kill_service and targeted 2.1 in 9901879