chore: Try to improve smoke test tracing to figure failure/flaky case causes [INS-5056]

[cwangsmv]

Current smoke test tracing logic has the following issues:

• Although we set trace mode to 'retain-on-failure', it actually captures traces for every test case, no matter sucess or failure. As a result, the trace file is too huge to navigator to failed cases easily.
• The tracing file for common flaky cases with 'timeout' problems does not contain screenshots. This made it difficult to triage root causes of the flaky cases.

Fix these issues with conditional tracing logic and fix two failed cases cause by the tracing change