Detect and distinguish dual-use libfuzzer args for fuzzing

[ranweiler]

When creating libfuzzer jobs, users may add -runs=N to target_options to try to handle situations like target code that leaks memory. This is fine and expected, and handled by the libfuzzer_fuzz task.

However, when a libFuzzer is invoked in test case repro mode, the -runs=N option says "test the given input(s) N times each". This is definitely not what we want in analysis tasks like coverage, libfuzzer_crash_report, &c.

Revisit our task arguments, and fix this end-user facing API. Make it easy (at the job template level) to clearly distinguish between libFuzzer args meant for fuzzing only, test case repro, or both. This is more easily avoided outside of templates, where tasks must be created individually.

[ranweiler]

This also impacts our current integration tests: https://github.com/microsoft/onefuzz/blob/main/src/integration-tests/integration-test.py#L107