[tests] Track evaluation time across performance runs

[jonathanpeppers]

Summary

• track the maximum MSBuild project evaluation duration across every Profile iteration
• apply the same aggregation to ProfileTask
• use the aggregate for slow-machine classification instead of only the final binlog

Fixes the valid finding in #12389 (comment).

Testing

• dotnet build tests\MSBuildDeviceIntegration\MSBuildDeviceIntegration.csproj --no-restore -v:minimal -p:JavaCPath='C:\Program Files\Microsoft\jdk-17.0.16.8-hotspot\bin\javac.exe' -p:JarPath='C:\Program Files\Microsoft\jdk-17.0.16.8-hotspot\bin\jar.exe' (succeeded; 16 existing warnings, 0 errors)

[Copilot]

Pull request overview

Updates the MSBuild device integration performance tests to track the maximum MSBuild project evaluation time across all iterations in a profiling run, and uses that aggregate when deciding whether to mark a failure as “slow CI machine”.

Changes:

• Track and aggregate max ProjectEvaluation duration across Profile() iterations.
• Apply the same max-evaluation aggregation to ProfileTask().
• Use aggregated max evaluation time for slow-machine classification instead of relying on only the final iteration’s binlog.

[jonathanpeppers]

@dalexsoto review

[dalexsoto]

The maximum evaluation duration is now aggregated across all Profile and ProfileTask iterations, so slow-machine classification no longer depends on only the final binlog.