[GR-71655] Improve error logging and detection in PEBenchmark.

[graalvmbot]

On compilation failure, PEBenchmark did not indicate a problem (other than reporting NaN for compilation timings), because the exception it throws gets thrown from a callback on a compiler thread.

This change:

• adds an explicit check for empty data points.
• updates listener callbacks to log errors to stderr; adds an mx benchmark failurePattern to check for error logs.
• adds a new gate-truffle-test-compiler-benchmarks-graalvm-ce-latest-linux-amd64 tier3 job to test for compilation failures.