fix: 7D/30D chart shows 0.0s on latency benches

[Flotapponnier]

Summary

On benches with unit: s (e.g. aggregator-head-lag), switching to the 7D or 30D time range showed all Y-axis labels and tooltip values as "0.0 s" despite the chart lines rendering at clearly different heights.

Root cause: double division by 1000.

• The blob stores latency values in milliseconds (OCB internal convention for unit: s)
• /api/series was dividing values by 1000 before sending — originally added for the video renderer, which expects seconds and appends the unit label directly
• The time-series chart received values already in seconds, then fmtUnit(value, "s") divided by 1000 again → 0.2 s became 0.0002 s → displayed as "0.0 s"

The same double-conversion existed for bps benches (divided by 100 in API → fmtUnit divides by 100 again).

Fix: add ?raw=1 query param to /api/series that bypasses all unit conversions. The chart always passes raw=1 (values stay in the internal ms/bp convention that fmtUnit expects). The video renderer omits raw=1 and keeps receiving display-friendly seconds/percent values.

Files changed

• src/app/api/series/[slug]/route.ts — ?raw=1 skips the / 1000 and / 100 conversions
• src/components/time-series-chart/index.tsx — all 3 lazy-fetch buildQs helpers now include raw: "1"

Test plan

• aggregator-head-lag?chain=base → 7D and 30D tabs should now show real values (e.g. "0.2 s", "1.3 s") not "0.0 s"
• Same for any other unit: s bench with 7D/30D data
• Video renderer still works (it doesn't pass raw=1)
• 1H / 24H tabs unaffected (they use benchmark.extras.series24h directly, not this API)

🤖 Generated with Claude Code