Add /wpt/runs page with the latest completed WPT run per browser

[nicoburns]

Summary

Adds a /wpt/runs page, similar to https://wpt.fyi/runs. It shows the latest wpt.fyi master run for each browser on the WPT comparison page, newest first by when it was added to wpt.fyi (created_at). A "Latest runs" link sits before "Focus areas" on /wpt.

• Channels: reuses wpt_compare::PRODUCTS minus HIDDEN_PRODUCTS (Chrome Canary, Firefox Nightly, Safari TP, Ladybird, Servo), so the two pages can't drift apart.
• Request: a single api/runs?label=master&max-count=5&product=…&product=… request, where max-count applies to each product. aligned isn't set. The latest run is picked by created_at rather than time_start, because a run that started earlier can finish later.
• Columns: browser, with its version underneath, and a link to the results; WPT commit; start time; run duration; time added to wpt.fyi; and upload delay (created_at − time_end).
• Refreshing:
  • A background task started in main refreshes WPT_RUNS_CACHE on startup and then every hour, whether or not anyone visits.
  • Page loads also refresh it in the background once the data is more than 5 minutes old, via get_or_refresh(FRESH_FOR, Duration::MAX). Visitors never wait on a refresh.
  • load_wpt_runs only ever runs one refresh at a time, using a module-level REFRESH_LOCK.
• "Refreshing data…" indicator: shown next to "Updated N ago" when refreshing is true:

  refreshing: wpt_runs::is_refreshing() || entry.cached_at.elapsed() > FRESH_FOR

  is_refreshing() checks whether REFRESH_LOCK is held. The stale check covers the moment just after get_or_refresh starts a background refresh, before that refresh has taken the lock.

In-progress runs, which need GITHUB_TOKEN, are split out into #34, which is stacked on top of this PR.

Testing

• cargo clippy --all-targets: no new warnings. The only warning is the existing one in src/downloads.rs.
• Ran locally and loaded /wpt/runs: 5 rows from live wpt.fyi data. A cold load took about 0.46s. After a restart, the first request was served from cache in about 1ms ("Updated just now") because of the startup refresh.
• Tested the indicator in a temporary build with FRESH_FOR set to 3s. Loads of stale data showed "Refreshing data…", and it disappeared once the refresh finished.

Link to Devin session: https://dioxus.staging.devinenterprise.com/sessions/94bf6cad086a43818116ad129df06999
Open in Devin Desktop: https://dioxus.staging.devinenterprise.com/desktop/session/94bf6cad086a43818116ad129df06999?variant=devin-insiders
Requested by: @nicoburns

[staging-devin-ai-integration]

I'll fix CI failures and address comments from users with write access that start with 'Devin'.

• Disable automatic comment, CI, and merge conflict monitoring