Updated dictionary merges - need a better tracking system for subtrees.

messages if you get the names wrong.  Contact me if the error messages are not

helpful enough to solve your problems, or if you have new ideas on functionality.

Substitutions

-------------

Before `make.py` runs jobs, it generates jobscripts from your yaml.

The `environment`, `script` and `retry_script` variables of each job

go through substitution at creation time for each script.

In the case of a `Restartable` jobtype, the `retry_script`

and `complete` tests are also substituted.

Specifically, the python format() function is invoked.

This means ordinary brackets, `{}`, need to be written as

double-brackets, `{{}}`.  Also, it means you can use the

variables defined in the yaml itself in the script.

In addition to those variables, you also have access to

everything defined the `reqs` dictionary of jobtypes.py.

Finally, every project also has:

    proj = the project ID input to `make.py`

    jobname = the job name itself (top-level header for the job)

        # Fill-in from kws and type-check class vars

        for var, cls in self.reqs.items():

            if var in kws: # initialize from yaml data

                val = kws[var]

                val = kws[var].copy()

                if isinstance(val, dict) and hasattr(self, var) \

                         and isinstance(getattr(self,var), dict): # merge together

                   dict_merge( getattr(self,var), val ) # recursively merge parameter hierarchies

   job0 = types[jobtype]

   if issubclass(job0, types['Chain']):

       for i,d in enumerate(args['chain']):

            if isinstance(d, tuple): # already processed.

              continue

            assert len(d) == 1

            for type1, args1 in d.items():

                pass