Rework custom test machinery to give the individual tests more control

on how they're run, and combine environment.json and check.py

test/02_test/cibuildwheel_test.py

@@ -0,0 +1,23 @@
+import subprocess, sys, os
+from glob import glob
+project_dir = os.path.dirname(__file__)
+sys.path.append(os.path.join(os.path.dirname(project_dir), 'shared'))
+
+import utils
+
+def test():
+    # set up the environment
+    env = os.environ.copy()
+    env["CIBW_TEST_REQUIRES"] = "nose",
+    # the 'false ||' bit is to ensure this command runs in a shell on
+    # mac/linux.
+    env["CIBW_TEST_COMMAND"] = "false || nosetests {project}/test",
+    env["CIBW_TEST_COMMAND_WINDOWS"] = "nosetests {project}/test"
+
+    # build & test the wheels
+    subprocess.check_call([sys.executable, '-m', 'cibuildwheel', project_dir], env=env)
+    
+    # also check that we got the right number of built wheels
+    expected_identifiers = utils.cibuildwheel_get_build_identifiers(project_dir)
+    built_wheels = glob('wheelhouse/*.whl')
+    assert len(built_wheels) == len(expected_identifiers)