Fix personal nodriver evaluate normalization

[zy6p]

Summary

• always normalize nodriver tab.evaluate() results in personal captcha mode
• execute fingerprint extraction JS as an IIFE so the object is actually returned
• fixes personal score-test DOM parsing and restores full browser fingerprint extraction

Reproduction

I reproduced this against current upstream main locally with BrowserCaptchaService.get_custom_score() on https://antcpt.com/score_detector/.

Before this patch:

• token acquisition succeeded
• verify_result ended as success=false with source=antcpt_dom_timeout
• get_last_fingerprint() only had the fallback UA/lang because the full object was not parsed

Root cause:

• _tab_evaluate() only normalized results when return_by_value=True, but most callers do not pass that flag and receive nodriver's serialized object-entry structure instead of plain Python values
• _extract_tab_fingerprint() passed an uninvoked arrow function (() => { ... }) so nodriver did not return the intended object payload

After this patch:

• verify_result.success=true
• score=0.9
• full fingerprint fields are returned (user_agent, accept_language, sec_ch_ua, sec_ch_ua_mobile, sec_ch_ua_platform)

Validation

• python3 -m py_compile src/services/browser_captcha_personal.py
• local personal get_custom_score() reproduction before/after patch on upstream main

[anthony9981]

Remove return_by_value check is not good, change default value above to True is better

[zy6p]

Addressed in 8a8f5c4: I kept the return_by_value switch, changed its default to true, and only normalize when return_by_value is enabled. This preserves the non-value path while making existing evaluate callers return plain Python values by default.

[zy6p]

Updated per review: _tab_evaluate now keeps the return_by_value switch, but defaults it to true so the nodriver evaluate result is normalized by default for existing internal callers. The non-value path is still preserved if it is needed later.

[genz27]

已同步 main，冲突已解决，并完成本地测试验证。