Rebuild the site around the benchmark comparison

[martian56]

Rebuilds web/ on top of quanta-port, so the site can describe the native
protocol instead of the PostgreSQL wire protocol that branch removes.

What changed

The page opens on the four-engine comparison rather than a claim. The table
sits in the first viewport at full width with its commit, date and container
limits attached, and the workloads QuantaDB loses are drawn at the same
weight as the ones it wins. PostgreSQL's write lead is marked, not buried.

Structure varies section to section instead of repeating one rig: the field
table, the five-machine matrix, a prose pair on the engine, two code blocks,
and the milestone list.

The tables are CI output now

readme_bench.py gained a second target. It rewrites the marked blocks in
web/index.html from the same result artifacts it already uses for the
readme, so the published numbers cannot drift apart, and bench.yml commits
both files. Verified idempotent: running the publisher twice against fixture
data produces identical output.

The five-machine table is now ordered by write throughput, which puts the
Windows runner and its 540 ops/s at the bottom where the callout reads
correctly. A profile is flagged when its writes fall below a quarter of the
best profile's or its write tail passes 20 ms.

Protocol copy

No psql quickstart, no "Speaks PostgreSQL" card, no Postgres wire claim.
The quickstart builds from source, because there is still no published
image, and connects over TCP on 6626.

One thing reviewers should look at

The field numbers were produced by driving QuantaDB through its
PostgreSQL-compatible port with psycopg2, which is what field_bench.py
still does. That port does not exist on this branch. The measurements are
real and the engine path behind them is unchanged, but the comparison is not
reproducible against this commit until reference.yml and field_bench.py
are re-pointed at the native client. The page states this in place rather
than implying the run is current. That harness work is still open.

Supersedes #76.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Pro Plus

Run ID: 1a98bd27-9f53-42ac-bd2c-f24a774d5ea6

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch site-rebuild

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}