Add a perfstar baseline for tracking the impact of generating binlogs

[baronfel]

Summary

We want people to make binlogs all the time, but we should be clear-eyed about the cost of doing so. We should make a PerfStar baseline that tracks the cost of generating binlogs for a representative project (orchard core) on stable hardware to give users a way of thinking about this cost, as well as giving ourselves a target to minimize.

Background and Motivation

The latest orchard code commit on my machine has a ~25% overhead for binlog generation. I

• cloned the repo
• ran dotnet restore to pre-load the nuget caches
• triggered the build without binlogs via dotnet build
• run git clean -xdf to remove intermediates
• ran dotnet build-server shutdown to kill the persistent build servers/nodes
• triggered the build with binlogs via dotnet build -bl

and that gave me timings of 105.6s and 135.0s respectively on my machine (+- a tiny bit of jitter across iterations). This is on an AMD Ryzen 7 5700x3d, 64 GB of decently-fast RAM, and an NVMe SSD.

Proposed Feature

We should make a harness that does the above in a more repeatable way and tracks the ratio over time.

Alternative Designs

No response

[maridematte]

This is one of the idea for the next perfStar improvements (you can find it in the one pager, which I'll link once the PR is completed).