Logging SDK speed regression

[emilk]

The C++ logging SDK is adding a lot of overhead, and we need to figure out why.

Reportedly 2ec5ce7 was much faster than 0.10.0, so something in 2ec5ce7...0.10.0 broke (TODO: verify).
Maybe c0521ee causes a Debug build of arrow-cpp to be built or something?

$ time RERUN=on ./build/rerun_vrs_example sample.vrs
real    0m15.105s
user    0m16.409s
sys     0m2.981s

$ time RERUN=off ./build/rerun_vrs_example sample.vrs
real    0m1.060s
user    0m1.043s
sys     0m0.017s

More info: https://rerunio.slack.com/archives/C04UBTMMGJK/p1698696318766679

[jleibs]

Using -DRERUN_DOWNLOAD_AND_BUILD_ARROW=off definitely seems to alleviate the problem.

With: -DRERUN_DOWNLOAD_AND_BUILD_ARROW=on

$ time RERUN=on ./build/rerun_vrs_example sample.vrs
real    0m15.105s
user    0m16.409s
sys     0m2.981s

With: -DRERUN_DOWNLOAD_AND_BUILD_ARROW=off

$ time RERUN=on ./build/rerun_vrs_example sample.vrs
real    0m3.643s
user    0m4.754s
sys     0m1.643s

Note, there's still a fair bit of overhead relative to having logging off all together:

$ time RERUN=off ./build/rerun_vrs_example sample.vrs
real    0m0.973s
user    0m0.952s
sys     0m0.022s