First step to simplify execKarma method, which is overcomplicated and has too many responsibilities.

…gic (#3493)

This is a bit bigger change, but modifications are closely related and I think it makes sense to land them all together.

The main change is in test/e2e/support/world.js file, where all important paths are defined:

- `workDir` is the current working directory for the tested Karma process - `test/e2e/support`. Same as before.
- `sandboxDir` moved from `tmp/sandbox` in the repository root to `test/e2e/support/sandbox`. Sandbox directory is reset before every scenario (see `test/e2e/step_definitions/hooks.js`) and this is where all generated files should be (hence change of paths in `browser_console.feature`). This also makes things cleaner as now all paths in the .feature files are relative to `workDir`.
- `configFile` is path where Karma config file is generated. It no longer contains hash as it was not really needed. It may be re-introduced later if tests are run in parallel, but then it makes more sense to have it as a part of `sandboxDir` instead.
- `karmaExecutable` is an absolute path to the Karma executable.

The change from `__dirname + '/` to `_resolve('` in two .feature files is to avoid having two different hacks to resolve absolute paths.

The last change in this file is to simplify config generation methods.

Remaining changes in `test/e2e/step_definitions/core_steps.js` are adapting steps to the above changes by removing bunch of boilerplate, which is no longer necessary.

From the description of the steps it makes sense to write it straight away. Also this is more flexible and allows to introduce steps, which do not require config file in the future PR.

The existing code had pretty confusing logic (old lines 33-36)`: when background server start fails it is actually considered a success, but child process is not started. Because of this `lastRun` contains result output of the `karma start`. This is no longer the case, hence two test cases had to be updated as they never executed `karma run` despite the statement and were asserting against `karma start` output. This should be more clear now.

As background server process now has its own variable to store output, there is no need for a dedicated runOut command and it can be removed.

Finally!

Introduced extra step "I wait until server output contains" instead of hard-coded timeout. It will regularly evaluate a condition until it fulfills. This approach should potentially perform a little bit better than the previous solution.

Also generalize a step to accept any CLI arguments instead of only log-level. This should make this step more flexible. The removed step was not very clear as to which command it applies to (foreground or background) and is fully covered with expanded "I {command} Karma ..." step.

These were lost after migration to yargs as CLI parser.

Fixes #3474

## [5.0.5](v5.0.4...v5.0.5) (2020-05-07)

### Bug Fixes

* **cli:** restore command line help contents ([#3502](#3502)) ([e99da31](e99da31)), closes [#3474](#3474)