Stop using m_summary in the global globals for telemetry

[florelis]

We were observing crashes in TelemetryTraceLogger::LogFailure() due to heap corruption. The root cause of this is that for many types of COM calls we don't have a thread local telemetry trace logger, but instead use a process global. If there are multiple simultaneous calls, there is a possibility for a race condition, specifically around writing to the TelemetryTraceLogger's m_summary.

The long term solution would be to stop using a process globals there, but that is tricky. As a band-aid to prevent these crashes, disable the use of m_summary in the process globals.

Tested by creating a local repro of the crash and verifying it didn't crash after the change. Repro'd by adding an artificial THROW_HR_MSG to force going to LogFailure() and then using the sample COM caller project to generate many simultaneous calls that would hit it.

Credit to @JohnMcPMS for figuring out the issue and the fix :)

Related to #1880

Microsoft Reviewers: Open in CodeFlow

[JohnMcPMS]

I only guessed at the cause, but this is definitely a potential issue. No guarantees that this fixed anything that is actually happening 😉

[yao-msft]