feat: export build_launch_options helper for composing with existing Playwright instance

[mvanhorn]

Summary

Users who run their own sync_playwright() context can now pull CloakBrowser's launch args into it with one call instead of copying the body of launch() each release.

Why this matters

In #153, @chrisemke described maintaining scrapers that share a single Playwright instance across multiple stealth backends, and asked whether launch() could accept a playwright param or expose a lower-level helper. The maintainer reply asked the same question.

This PR takes the lower-level-helper route to keep the existing launch() signature stable. Adding a playwright= kwarg can layer on later if preferred - it composes with this helper.

Testing

• 124 unit tests pass: test_build_args, test_config, test_proxy, test_backend, test_geoip, test_extension_loading.
• py_compile clean across cloakbrowser/browser.py, cloakbrowser/__init__.py, examples/external_playwright.py.
• build_launch_options() returns executable_path, headless, args, ignore_default_args; adds a proxy dict only when a proxy URL is passed.
• launch() is refactored to delegate to build_launch_options(), so the existing call path produces identical kwargs for pw.chromium.launch().
• The example at examples/external_playwright.py shows the sync_playwright() + build_launch_options() pattern from the maintainer reply.

Scope kept to the sync path. Async launch_async, launch_context, launch_persistent_context are untouched and can follow as a separate PR if useful.

Fixes #153

[Cloak-HQ]

Thanks for the PR, @mvanhorn! This addresses #153 nicely.

Two things before we can merge:

1. Missing humanize_browser() helper — The JS equivalent (feat(js): export composable launch helpers #244) exports both buildLaunchOptions and humanizeBrowser. For parity, this should also export a humanize_browser(browser, ...) so users can apply humanize to their own Playwright instance.

2. launch_async() still has duplicated logic — launch() now delegates to build_launch_options(), but launch_async() still has its own copy of the same arg-building code. It should delegate to build_launch_options() too, otherwise they'll diverge over time.

Let us know if you'd like to update, or we can handle it on our end.

[mvanhorn]

Thanks for the careful review. Both items addressed in 17c85c3:

1. Exported humanize_browser() and humanize_browser_async(). Callers who launched their own Playwright instance (e.g. via build_launch_options()) can now apply the stealth humanize patches with humanize_browser(browser, preset='default', config=None). The sync and async variants stay split because patch_browser and patch_browser_async are themselves separate. launch(humanize=True) and launch_async(humanize=True) now delegate to these helpers, so the patch logic stays single-sourced.

2. launch_async() delegates to build_launch_options(). The duplicated geoip/proxy/webrtc/build_args block is gone from launch_async. It now mirrors launch() exactly: build the opts dict, log args count, await pw.chromium.launch(**opts). Dropped the # noqa: C901 since complexity is back under threshold.

3. Tests. Added tests/test_build_launch_options.py covering: build_launch_options schema (executable_path, ignore_default_args, timezone/locale flags merged into args), both humanize helpers (with patch_browser / patch_browser_async mocked so the test doesn't need a real Browser), and launch_async delegation (asserts build_launch_options is called with forwarded kwargs and pw.chromium.launch is awaited with the returned dict). Focused suite passes; full pytest requires playwright installed which I didn't add as a dev dep here.

Let me know if you'd prefer a single humanize_browser() that auto-detects sync vs async by inspecting the Browser object instead of the two-name form; happy to switch.

[mvanhorn]

Both done, and rebased onto main (force-pushed).

1. Added humanize_browser() (and humanize_browser_async() for the async path) for parity with the JS side - they apply the same humanize patching launch(humanize=True) does to a user-supplied browser. Both re-exported from the package.
2. launch() and launch_async() now both delegate to build_launch_options(), so their arg-building can't drift. I also folded the current launch() behavior (license_key/browser_version, start_maximized, the Windows-font warning, and env handling) into build_launch_options() so delegating doesn't drop any of it.

Verified: the build_launch_options + delegation tests pass, including a new one asserting launch_async delegates. The only failing tests locally are the real-browser bot-detection ones that need the licensed binary, and those fail the same way on main.