runtests: start eval-test first because it's the long pole

Running with 8-way concurrency on my desktop, and after the
optimizations made in the last few commits, the whole run finishes
when this one test does, long after all other tests have completed.
Clearly it ought to be broken up; first and easier, do the best we can
by putting it right at the head of the queue.

Author: gnprice

runtests.py

@@ -200,11 +200,16 @@ def add_myunit(driver: Driver) -> None:
         elif mod == 'mypy.test.testpythoneval':
             # Run Python evaluation integration tests separetely since they are much slower
             # than proper unit tests.
-            driver.add_python_mod('eval-test %s' % mod, 'mypy.myunit', '-m', mod, *driver.arglist)
+            pass
         else:
             driver.add_python_mod('unit-test %s' % mod, 'mypy.myunit', '-m', mod, *driver.arglist)
 
 
+def add_pythoneval(driver: Driver) -> None:
+    driver.add_python_mod('eval-test', 'mypy.myunit',
+                          '-m', 'mypy.test.testpythoneval', *driver.arglist)
+
+
 def add_stubs(driver: Driver) -> None:
     # We only test each module in the one version mypy prefers to find.
     # TODO: test stubs for other versions, especially Python 2 stubs.
@@ -349,6 +354,7 @@ def main() -> None:
     driver.prepend_path('PYTHONPATH', [driver.cwd])
     driver.prepend_path('PYTHONPATH', [join(driver.cwd, 'lib-typing', v) for v in driver.versions])
 
+    add_pythoneval(driver)
     add_basic(driver)
     add_selftypecheck(driver)
     add_myunit(driver)