## Test subcommand

Adds a test subcommand through py.test (with ~all py.test options forwarded directly):
- flexible testing specification, the subcommand can take Odoo modules, directories and/or files as positional parameters and will apply [py.test's discovery conventions](http://pytest.org/latest/goodpractises.html#test-discovery) to fetch Python test cases, "data" test cases will be extracted from a module's `__openerp__.py` if that matches other filters
- _only runs selected tests_, if requesting the testing of `product` all its dependencies (and auto_install dependents) are installed (with demo data) but only the tests of `product` are run
- with no positional parameters, just starts looking for tests in the current working directory
- automatically creates and deletes a temporary db if none is provided, if one is provided does not delete it (but creates it if it's missing) which can allow reusing dbs
- uses an inited (but not specifically configured) Odoo logger and a pytest in default display mode (line per test file, captures stdout/stderr)
- recap at the end of testing with display of failed test stack traces & captured stderr/stdout
- provided marks to ignore categories of tests, only current mark is `http` so `odoo.py test -m 'not http'` will not start an http server and won't run any HTTP test
- existing "unittest" tests mostly work as-is (unless they were mucking with the registry or even runtest protocol), alternatively can write lighter "pytest" tests
- fixtures make for pretty clean setup/teardown & composition
- slightly to greatly faster tests depending on the selected module(s)
- future improvements via existing addons (e.g. [pep8](https://pypi.python.org/pypi/pytest-pep8), [coverage](https://pypi.python.org/pypi/pytest-cover)) or more formal results reporting (e.g. [junitxml result logs](http://pytest.org/latest/usage.html#creating-junitxml-format-files)) or new addons for e.g. better JS or tour integration

e.g. `odoo.py test mail account` (will run all non-skipped tests in the mail and account addons), `odoo.py test  addons/product/tests/test_pricelist.py` (will just run the two test cases in test_pricelist.py), `odoo.py test --pdb -k test_20_pricelist_uom` (will only run that test case, should drop into pdb if the test fails)
## `--test-enable`
- most of the registry-installation functions moved to methods of Registry
- Registry.load_modules changed to a coroutine and testing part moved outside of it to RegistryManager.new, which external runner does _not_ use see 3219863
- "on-install" testing uses same test discovery subcommand does with custom test reporter
  - reporting not identical to old reporting but pretty close (and improvement in some ways)
  - replaced old assertion_reporter (which just counted) by normal assertion errors in XML and YAML
  - all failed tests summarised/printed at the end of installation, no need to go hunt in the log
- remains compatible with runbot (I think, went into runbot source to see how that worked)
## other changes
- some tests were implicitly skipped (weren't imported by test package), changed to explicit skipping
- various imports improved, some of them may not be necessary after all. May be useful if we eventually try to make just calling `py.test` directly possible
- moved and removed imports so pytest and unittest2 are only imported when actually trying to _run_ tests
- removed `--test-file` behaviour, superseded by `odoo.py test <testfile>`

/cc @odony @antonylesuisse @tde-banana-odoo