Add TraceInfo to dotnet-trace traces (similar to PerfView)

[felipepessoto]

When we capture a trace with PerfView, it collect some useful trace information, like timestamps from start and end.

I guess Perfview uses "Windows Kernel/EventTrace" event. I think we can find a cross-plat alternative for dotnet-trace.

Docs: https://github.com/microsoft/perfview/blob/master/src/TraceEvent/EventPipe/NetPerfFormat.md
https://github.com/microsoft/perfview/blob/master/src/TraceEvent/EventPipe/EventPipeFormat.md

[noahfalk]

Do you care about this from the angle of what is displayed in the PerfView UI, or specific data that is present in the trace because you will be parsing it yourself?

Right now every event in the nettrace format has a timestamp on it and every block of events contains a timestamp range on it. Other options may be possible too.

[felipepessoto]

@noahfalk at first I needed it to show the event counters with the absolute date/time instead of msec since the beginning of the trace: microsoft/perfview#1013

Brian already confirmed that we had the start time in the trace headers so I can calculate the time of each event counter sample.
Anyway, these information (machine name, cpus, utc offset, etc) are always useful when troubleshooting issues in customer environment with several server:

[noahfalk]

Thanks @felipepessoto ! We'll probably need to do a more thorough analysis but off the cuff I think this is the current state we are working from - all the missing/inaccurate pieces are what this bug would need to address:

MachineName, OS, OS Build Number - missing
UTC offset and analysis/collection delta - missing
OS Boot time - missing
Trace start time - available
Trace end time - can be calculated by scaning each EventBlock and taking the max of the EventBlock header timestamps. We can't place the value in a header because the format is a streaming format and we don't know the final timestamp until the end of the file.
CPU Freq - present in the trace but I think the data written there isn't accurate
NumberOfProcessors - available
MemorySize - missing
Pointer Size - available
Sample profile interval - not sure if it is accurate
Total events - calculatable by scaning the file
Lost events - calculatable by scanning the file
File size - available from the OS, assuming the session is a file based session rather than a streaming session