Several fixes for existing kernels

[jprotze]

This is the artifact for an HPC Bugfest submission. The PR can be split into parts for review, if necessary.

This PR fixes several issues with current kernels:

• exhaustive execution time for some kernels (infinite loop in a data-race-free kernel ?!?)
• kernels without data race labeled with yes
• result for some kernels depends on random input or scheduling decisions, making the classification result random/non-reproducible, especially for tasking kernels
• -> introduce variants of these benchmarks that explicitly trigger the different choices

[chunhualiao]

@jprotze Could you please split this pull request into several smaller, independent ones?