In place of pulling out the list via the (now gone) algorithms()
method and going through it up at this level.
There is a slight difference in behaviour, in that the old way returned
the newest algorithm that HADN'T been executed already. However,
since the dialog is created immediately after the algorithm has been,
I don't see how that can be a problem.

This is better than pulling out the algorithm list and manipulating
things up the top.

There's no longer any multi-threaded execution going on, so I think
this test will behave now.

I found a better way to mimic a long running algorithm rather than
actually having one - just have it always say it's still running
if asked. This is just as good for testing this class as the old way.

We can do essentially the same testing via other public methods.

This was the place where there was previously a race condition.
The iterators in the list returned by the algorithms() method
could be invalidated by another thread creating more algorithms.

...of a given name.
N.B. Had to temporarily comment out some tests that used the removed
algorithms() method.

This breaks stuff. Fixing it is the next few steps....

It does not currently handle properties that don't exist when the algorithm is
initialized very well.

Conflicts:
	Code/Mantid/Build/wiki_tools.py
Re #6630. Merging master onto the branch because of a conflict other
people have had.