fix(browser): normalize wait timeouts

[hiSandog]

Summary

• normalize \ in both daemon and direct CDP paths so adapters can pass seconds while older millisecond-style callers keep working
• switch \ from a fixed sleep to \ plus a selector wait, matching the newer intercept semantics
• make \ use an explicit 10-second selector timeout and add regression coverage for the new wait helper and 36kr article flow

Testing

• \

@jackwener/opencli@1.5.5 test
vitest run --project unit src/browser/wait.test.ts src/browser/dom-helpers.test.ts src/browser/cdp.test.ts

RUN v4.1.1 /Users/sandog/sandog/code/kaiyuan-space/opencli

Test Files 3 passed (3)
Tests 13 passed (13)
Start at 14:46:42
Duration 349ms (transform 82ms, setup 0ms, import 113ms, tests 259ms, environment 0ms)

• \

@jackwener/opencli@1.5.5 test:adapter
vitest run --project adapter src/clis/36kr/article.test.ts src/clis/36kr/hot.test.ts

RUN v4.1.1 /Users/sandog/sandog/code/kaiyuan-space/opencli

Test Files 2 passed (2)
Tests 4 passed (4)
Start at 14:46:42
Duration 79ms (transform 36ms, setup 0ms, import 42ms, tests 10ms, environment 0ms)

• \

@jackwener/opencli@1.5.5 typecheck
tsc --noEmit

[Astro-Han]

The 36kr wait improvement and the sinafinance fix both make sense. One thought on the approach:

WaitOptions.timeout is already defined as seconds, and only sinafinance/rolling-news.ts passes the wrong unit (10000 instead of 10). The simpler fix would be to just correct that one caller rather than introducing a unit-guessing function. Auto-detecting "is this seconds or milliseconds?" based on a >= 1000 threshold adds an implicit convention that could confuse future contributors — someone passing timeout: 1000 (1000 seconds) would silently get 1 second instead.

Enforcing one consistent unit (seconds, as already documented) and fixing the one wrong call site keeps the codebase simpler.

[hiSandog]

Good point — I agree keeping a single consistent unit (seconds) is simpler and less error-prone. I reverted the normalizeWaitTimeoutMs helper entirely and kept the original (timeout ?? default) * 1000 pattern. The only functional changes are:

• 36kr/article: improved wait to use waitForCapture + selector wait instead of a blind sleep
• sinafinance/rolling-news: fixed the wrong unit from 10000 (ms) → 10 (s)

Tests pass. Thanks for the review!