DuckDB backend for benchmarks website

[AdamGS]

Summary

This PR started as a joke with @joseph-isaacs, but I actually think the result is quite nice (even considering the diff size).

This PR changes the benchmarks website backend from having a specialized ETL in JS, to processing the raw JSON into DuckDB tables, and using SQL to serve all endpoints.

Tooling

Moving to SQL also makes it easier to explore data locally, and this change includes tools and documentation on how to get the data into a local DuckDB instance, including generating all required SQL statements and downloading the data.

Docs

This PR includes new docs about the benchmarks website:

• README.md includes general setup, development and how to explore data locally.
• SCHEMA.md describes the various tables and views, and how the API uses them
• ETL.md attempts at describing the full lifecycle of the data, from being generated by the post-merge CI run, how various fields actually look and the processing steps it goes through before reaching the API.

[codspeed-hq]

Merging this PR will not alter performance

✅ 1163 untouched benchmarks
⏩ 1457 skipped benchmarks¹

---

_{Comparing adamg/website-duckdb (bac763c) with develop (6ac69e7)}

Footnotes

1. 1457 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[joseph-isaacs]

Does this help with filter over graph points?

[AdamGS]

This PR is mostly just switches the logic from JS to SQL with DuckDB, the thing I realized though is that I would like to make the server more generic in general and maybe reshape the data so creating new graphs/groups becomes easier (hopefully).

[AdamGS]

my browser was just angry at this

[palaska]

the white logo should have transparent background, I think claude might be able to fix the svg

[AdamGS]

I think it does? It was just angry that it was srcset and not srcSet

[AdamGS]

I actually think this is a really good step, the follow ups are:

1. Normalize the raw data into more columns/fields.
2. Benchmarks report actual raw data, instead of rations or any processing.
3. Figure out a way to make the layouts and charts just config

[robert3005]

You’re going to make yet another charting tool by converting charts to config. You’d think there’s something for it already?

[AdamGS]

yeah I sort of hope to find something minimal here eventually, I think that's currently a stretch goal, getting nicer data and easier exploration seems much more important