Use the same API for convolution models as other models.

It turns out that the load_xsXXX() routines aren't necessary, since
the models can be created using the same interface as the other models
(after this change is made): e.g.

  ui.set_source(ui.xsphabs.gal * ui.xscflux.mdl(ui.xspowerlaw.pl))

This commit leaves in the load_xsXXX() routines, but they will probably
be removed in the next commit.

sherpa/astro/ui/__init__.py

@@ -48,7 +48,9 @@
 if hasattr(sherpa.astro, 'xspec'):
     _session._add_model_types(sherpa.astro.xspec,
                               (sherpa.astro.xspec.XSAdditiveModel,
-                               sherpa.astro.xspec.XSMultiplicativeModel))
+                               sherpa.astro.xspec.XSMultiplicativeModel,
+                               sherpa.astro.xspec.XSConvolutionKernel)
+                              )
 
     # Perhaps everything exported by the xspec module should be exported
     # here?
@@ -64,6 +66,8 @@
     # names, to make it easier to support different versions of
     # the XSpec library.
     #
+    # Is this still needed now that the models are added directly?
+    #
     for n in [n for n in dir(sherpa.astro.xspec) if
               n.startswith('load_xs') and n != 'load_xsconvolve']:
         globals()[n] = getattr(sherpa.astro.xspec, n)

sherpa/astro/xspec/tests/test_xspec.py

@@ -164,7 +164,8 @@ def test_convolution_models(self):
         # (log 10 of this is -8.8).
         lflux = -5.0
         #xs.load_xscflux("cmdl")
-        ui.load_xscflux("cmdl")
+        #ui.load_xscflux("cmdl")
+        ui.create_model_component('xscflux', 'cmdl')
 
         # If the test is run directly, this is not needed (i.e. the
         # variable cmdl is defined), but if run via 'python setup.py test'