[CompilerPerf] [WIP] trial: suppress needless uses of callvirt

[dsyme]

PR to reassess the perf impact of using call instead of callvirt in some situations. See context at #2844 and #468

[dsyme]

Well, using this flag at least doesn't seem to hurt performance (we wouldn't expect it to though), see https://github.com/Microsoft/visualfsharp/blob/master/tests/scripts/compiler-perf-results.txt#L76

startup-to-parseonly parseonly-to-check   check-to-nooptimize  check-to-optimize    nooptimize-to-debug
10.75                28.57                42.86                56.34                52.50

v. some recent times on master:

10.12                30.98                45.96                61.91                51.16               
10.31                30.51                46.54                62.74                51.05               

That may even indicate an improvement, e.g. three of the phases seem to improve - but note that the times are pretty flaky - regular +/- 10% - so each are within the margin of error. I'll do more test runs.

[KevinRansom]

@dsyme
What do you plan to do with this?

[dsyme]

@KevinRansom I'm not sure yet. Let's leave it open and I'll get more perf data (the compiler perf scripts analyze the master branch plus any open PRs with [CompilerPerf] in the title)

[KevinRansom]

@dsyme ... do you want to revisit this PR?

[KevinRansom]

Actually looks like nothing much left of it.