Fix performance of stale image check

[Eric-Arellano]

Closes #2518.

We were before checking every image inside await Promise.all() and only logging results at the end. This is too slow to keep that much in memory, and it results in the experience feeling laggy to not get results until the very end.

Instead, it makes more sense to check one image at a time and eagerly report violations.

Other improvements in this PR:

• Use ripgrep rather than git grep. The downside is requiring users have ripgrep installed, but this is a manual script we run with npx tsx, so it should be fine.
• Log progress every 20 images
• Add an allowlist.

example run

❌ image is unused: public/images/migration/dedicated_queue_no_jobs1.jpg
❌ image is unused: public/images/migration/switcher-small2.png
❌ image is unused: public/images/guides/normal_queue_with_providers1.jpg
❌ image is unused: public/images/guides/paulibasis.png
⏳ Checked 20 / 3519 images
❌ image is unused: public/images/qiskit-ibm-runtime/noisy-sim-circuit.png
❌ image is unused: public/images/qiskit-ibm-runtime/noisy-sim-estimator-circuit.png
❌ image is unused: public/images/qiskit-ibm-runtime/noisy-sim-estimator-ideal.png
❌ image is unused: public/images/qiskit-ibm-runtime/noisy-sim-estimator-noisy.png
❌ image is unused: public/images/qiskit-ibm-runtime/noisy-sim-sampler-ideal.png
⏳ Checked 40 / 3519 images
❌ image is unused: public/images/qiskit-ibm-runtime/noisy-sim-sampler-noisy.png
❌ image is unused: public/images/api/qiskit-addon-cutting/qiskit_addon_cutting-instructions-1.svg

[frankharkins]

Thanks!

[frankharkins]

The await Promise.all approach should work fine here. There won't be a race condition with numFilesChecked since Node is single-threaded. I'd expect it to make a difference since there's a lot of time waiting for commands to run. Just move L27 -> L34 so there's no await between incrementing and logging.

[Eric-Arellano]

I don't think this would work well because we could trigger too many concurrent calls to ripgrep, which can peg the system.