1.  NaN checking and terminate (final samples if pending, save, execute only
    breakpoints thereafter)
1b. NaN checking on stochastic multi-path and continue later paths up to
    termination.
1c. Terminate simulation after x hours (final samples if pending, save,
    execute only breakpoints thereafter) -- DONE

2.  Check (and fix?) seeding on the scheduler.  Are the chosen seeds unique?
    Document.

3.  Nonlinear k-space operators.
	Works in all IP and EX algorithms, although it the nonlinear part
	itself is never done in the interaction picture.
	a. New nonlinearKOperators element (same features as koperators)
	b. Search for terms in CDATA, just like EX algorithm, with less checks
	   on arguments.
	c. Define underlying variables like EX and initialise them.
	d. Go_space code (like EX)
	e. Check and edit CDATA of the actual nonlinear operator definitions.
	   This might reuse lines from the original.
	
	<CDATA[
	blah
	blah
	N1[#]=f(k,#);
	blah
	N2[#]=g(k,#);
	]>
	
	dpsi_dt = blahdeblah + N1[psi*h(x)*conj(psi)] + N2[phi];

	f.  Documentation	
	
4.  IP picture checks should cause barfs if people don't read doco.

5.  Allow mixed spaces for calculation of moment groups.

6.  Allow more tags to be specified by arbitrary C-code. So far, <seed>, 
    <paths>, <interval> and <domain>  could, for instance, be specified by a
    const in the <globals> section, or by a command-line input argument.
    Some tags I (Andy) am interested in generalising are:
	
	<lattice> (in field, integrate, and sample), <samples>, 
	<tolerance> (and related cutoff). Other possibilities include the
	mean for Poissonian noise, but I've never used it...

    Some of these tags will introduce complications. eg checking that lattice
    is divisible by samples, etc, will need to be done at run time. On that
    note, why is this enforced in the adaptive algorithms?
	
7.  Remove check for <lattice> divided by <samples> is integer for adaptive
    <integrate> algorithms.
