Clean up task output message handling

[hjoliver]

Close #1604.

• Task ID is no longer prepended to output messages at the client end and sent across the network (not necessary as the receiving object in the server has the same ID).
• Task ID is now only prepended to output messages inside the data structure handed to the broker to match with prerequisites.
• Treat message outputs the same as standard outputs - TaskID prepended as per the previous bullet point.
• Deprecated "message offset" placeholders in message trigger strings (see below Clean up task output message handling #1761 (comment))
• As a consequence of the previous points, simplify output message handling in task_proxy.py by getting rid of a bunch of Task ID prepend/remove operations.
• Eliminates some annoying redundancy in the suite log: [task-id] ... task-id succeeded is now just [task-id] ...succeeded, etc.
• Rename some associated modules and classes (e.g. output.py:output to message_output.py:MessageOutput - as this class is only used for message outputs).

[hjoliver]

Re deprecation of cycle point offset placeholders in message trigger strings ... this:

[[[P1M]]]
   graph = foo[-P1M]:out1 => bar
...
[runtime]
   [[foo]]
       script = """
cylc message "files ready for bar at $(cylc cyclepoint --offset=P1M)"
"""
       [[[outputs]]]
             out1 = "output files ready for bar at [P1M]"

is now equivalent to this for all practical purposes:

[[[P1M]]]
   graph = foo[-P1M]:out1 => bar
...
[runtime]
   [[foo]]
       script = cylc message "files ready for bar"
       [[[outputs]]]
             out1 = "files ready for bar"

On master, we enforce the use of the offset placeholders in message output strings in order to distinguish the message strings emitted by successive instances of foo. But this was only necessary because we were not prepending task ID to message strings as is automatically done for the standard outputs (succeeded etc.). Wracking my brains to recall how this crazy situation came about, I believe it goes right back to cylc-2 when there was no dependency graph (and therefore no cycle point offset notation in the graph), so this was the way to implement inter-cycle triggering in the case of message triggers.

I've changed the documentation and examples to remove all references to in-string offset placeholders, but it's still supported in the code for backward compatibility.

[hjoliver]

This code was duplicated in two places.

[hjoliver]

@matthewrmshin - please review or re-assign. Test battery passes here.

[matthewrmshin]

./tests/message-triggers/03-new-style.t is failing due to timezone issue.

[matthewrmshin]

I don't think you need to declare global for constants.

[hjoliver]

ah, yeah, just for warned

[matthewrmshin]

Looking good. Some minor issues noted. Test OK (with and without site/user global configuration) in my environment apart from the time zone problem in the new test.

@arjclark please review 2.

[hjoliver]

All comments addressed.

[matthewrmshin]

Tests all good.

[arjclark]

Looks ok to me. Test battery passing in my environment.