[maulanawirawan]

Summary

In scrapling/engines/_browsers/_validators.py, one type alias named Seconds is used for parameters that are consumed in two different units:

param	annotation (before)	actual unit	consumption	docs
timeout	Seconds = 30000	milliseconds	set_default_navigation_timeout() / page.goto() (Playwright, ms)	docs/fetching/stealthy.md: "The timeout (milliseconds)… default is 30,000 ms"
wait	Seconds = 0	milliseconds	page.wait_for_timeout(params.wait) (_stealth.py)	docs/fetching/stealthy.md: "The time (milliseconds)"
retry_delay	Seconds = 1	seconds	time_sleep(self._config.retry_delay) (_stealth.py)	docs/fetching/stealthy.md: "Seconds to wait between retry attempts"

So the annotation stated wrong units for timeout and wait (30000 "Seconds" ≈ 8.3 hours), and wait=2 reads like 2 seconds but waits 2 ms.

The ms behavior itself is documented — the problem is only that the shared alias name contradicts it. This bit me in practice: passing timeout=60 while following the static Fetcher convention (timeout is documented in seconds there) yields Page.goto: Timeout 60ms exceeded instantly.

What this PR does

Introduces a Milliseconds alias and uses it for timeout / wait in PlaywrightConfig (and the _fetch_params dataclass). retry_delay stays on Seconds.

• No behavior change: both aliases carry the identical Annotated[float, Meta(ge=0)] msgspec constraint — validation, defaults, and everything downstream are untouched. This is purely making the type annotation tell the truth.
• Cross-fetcher unit unification (static Fetcher.timeout is documented in seconds vs browser fetchers in ms) would be a breaking change right after v0.5 — happy to open a discussion if you'd rather go that way, but the alias correction stands on its own either way.

Validation

• ruff check clean.
• Test suite locally (Windows, dev @ e5d1e36): 2210 passed; the 10 failures I saw fail identically on clean dev without this patch (local env: shell/session-state tests) — so zero regressions from this change.

AI disclosure (per AI_POLICY.md)

This patch was written primarily with Claude Code assistance (analysis + code); reviewed and submitted by me.