Fix and improve rt-tests result processing

Context

Investigation following user feedback uncovered issues in the current rt-tests processing.

In particular:
• cyclictest results currently do not appear to have meaningful latency thresholds configured.
• parse_rt_tests_results.py reports latency lines as passing, with the overall result primarily determined by the cyclictest return code.
• As a result, simply providing a maximum value would not be sufficient unless the parser also understands and evaluates the threshold.

Goal

Review and improve how KernelCI interprets rt-tests, particularly cyclictest, so that reported pass/fail results represent meaningful latency expectations.

Tasks / points to investigate
• Define how cyclictest thresholds should be configured.
• Update result parsing to evaluate those thresholds.
• Ensure failures are represented correctly in KernelCI.
• Review how invocation parameters and measured values are exposed.
• Determine how this should integrate with notifications and regression reporting.

Start by reading parse_rt_tests_results.py and tracing how cyclictest invocation parameters, latency lines, return codes, and measured values become results. Clarify the threshold configuration and evaluation rules with the existing rt-tests flow, then verify that meaningful pass/fail outcomes propagate to notifications and regression reporting.