Simple logging benchmarks for C++ & Rust

[Wumpf]

What

• Part of Report logging benchmarks for C++/Python/Rust to CI #4100

Implements SDK sided logging benchmark for C++ & Rust.
Kept as simple as possible, meant for whole-process profiling so we capture all sideeffects. This makes tthe data generation ('prepare') inside the benchmark apps ofc quite tricky as it has to be as fast as possible.
Additionally, both Rust & Cpp app expose a way to get more fine grained logging. Cpp does this via a simple profiler scope, Rust via Puffin/re_tracing.

Logging always happens to a memory recording.
Data is currently never passed in in the Rerun format

Contains the tree initial benchmarks we wanted to have:

• points3d_large_batch
  • Single batch of 50mio points (color, position, radius, single label)
• points3d_many_individual
  • 1mio individual points with different time stamp each (color, position, radius)
• image
  • log 4 different 16k x 16k RGBA8 images (4GiB of memory!)

Running instructions in main.rs & main.cpp!

Timings on my M1 Max in seconds
(tests are not perfectly comparable, they do not do the exact same thing. Prepare times are also slightly different and most significant in the _large_batch test)

• points3d_large_batch
  • C++: 0.94s
  • Rust: 1.34s
• points3d_many_individual
  • C++: 16.86s (⚠️ there's almost certainly some involuntary allocation going on there)
  • Rust: 2.75s
• image
  • C++: 3.11s
  • Rust: 1.10s

Missing

• Python version
• utility script for building, running and publishing data

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)
• I have tested demo.rerun.io (if applicable)
• The PR title and labels are set such as to maximize their usefulness for the next release's CHANGELOG

• PR Build Summary
• Docs preview
• Examples preview
• Recent benchmark results
• Wasm size tracking

[jleibs]

Nice! Very informative to have these.