send sccache logs to telemetry

[msarahan]

Utiilizes work from rapidsai/shared-workflows#281 and rapidsai/shared-actions#44 to capture more build statistics into the telemetry data.

Example of captured sccache data from RMM:

This data will of course be plot-able over time, which should be useful.

[msarahan]

Sccache data is present on the output of this PR:

CC @bdice @vyasr

[davidwendt]

I don't see the ninja logs being sent. Perhaps just a typo in the title? I got my hopes up there for a minute.

[msarahan]

Sorry @davidwendt for the false hopes. The approach here will extend to ninja logs very easily. I don't know how to parse ninja logs into something useful, though. Brad pointed me to https://github.com/rapidsai/cudf/blob/branch-25.04/cpp/scripts/sort_ninja_log.py, but I haven't had a chance to port that to the telemetry side. Am I reading it correctly that you want a mapping of some number of filenames to their compile times? Is it OK to have the list truncated to some small-ish number of files (10?) Cardinality explosion is my main concern with having this in the telemetry results.

[davidwendt]

... Am I reading it correctly that you want a mapping of some number of filenames to their compile times? Is it OK to have the list truncated to some small-ish number of files (10?) Cardinality explosion is my main concern with having this in the telemetry results.

Yes. If we could get at least the top 25 perhaps? I'm hoping the telemetry data can provide a history to see if compile-times change over time perhaps? I can help you with the sort_ninja_log.py utility as well.

[msarahan]

I linked this PR for its comments at https://github.com/rapidsai/build-infra/issues/142

I need to finish out the existing functionality, but I hope to return to adding this soon.