#1008: more minimal start-up dependencies

[benfitzpatrick]

This closes #1008.

This adds dependencies on R1 start-up/async tasks only at the first cycle
point of the dependent task.

@hjoliver, please review.

[arjclark]

@benfitzpatrick - is #747 relevant here?

[benfitzpatrick]

I think #747's problem will be OK because the start-up mechanism is different,
and they now behave like normal task dependencies after a bit of graph-hacking
at the beginning of the run.

[hjoliver]

Runs correctly, but graphing is broken if a start-up task is present: the general cycling gets missed.

[benfitzpatrick]

I keep creating these graphing problems - I think I'll write some "dot" file tests for
cylc graph.

[benfitzpatrick]

Runs correctly, but graphing is broken if a start-up task is present: the general cycling gets missed.

This looks OK to me for the r1_multi_start_back_comp_start_up suite - which
suite can reproduce it?

[hjoliver]

[scheduling]
    initial cycle time = 2014080800
    [[special tasks]]
        start-up = start
    [[dependencies]]
        [[[0]]]
            graph = start => foo

cylc graph SUITE 2014080800 2014081000, on master:

On your branch:

[hjoliver]

We need to get this show on the road, and this fixes the run time behaviour, so I'm merging this and referring the residual graphing problem back to the main issue #1008.