Listing tests (--collect-only and --dtest-print-tests-only) only lists tests that will run according to other arguments specified
This patch fixes behaviour for both `run_dtests.py` and `pytest`.
- Error handling for invalid parameter values / combinations is in a single place (`dtest_config.py`) and is executed before we actually traverse through the tests
- We exit with just a clean error message instead of tons of spam
- `run_dtests.sh` will not loose the exit code of `pytest` any more so we can clearly detect when test cases collection fails
- removed a bit of boilerplate code from `run_dtests.py`, e.g. what it did with xml processing is simply provided with `-q` argument of `pytest`
- tests filtering has been refactored to be cleaner
- fixed filtering of resource intensive tests and other tests (note that except for upgrade tests, we took care only about test method annotations - module level annotations were ignored for vnodes, no_vnodes, no_offheap_memtables and resource_intensive, ...)
- added meta_tests for the filtering and parsing exception handling
- added special parameter --metatests which is enough to run all the meta tests
- fixed Travis configuration so that it runs meta tests
Note that now `run_dtests.py` seems to be redundant. If we need it only for listing dtests, we can simply achieve exactly the same effect using `--collect-only -q --ignore=meta_tests` arguments for `pytest` instead of `--dtest-print-tests-only`, plus we need to filter output with `grep '.py::'` (in order to not include the summary line) and pipe stdout to the target file. It is now simplified so `run_dtests.sh` just uses `pytest` with those arguments.
patch by Jacek Lewandowski; reviewed by Tomek Ĺasica and Mick Semb Wever for CASSANDRA-16399