Python: capture the native element tree per action in trace mode #570

What

On a native Appium session, the Python adapter's trace actions carry a screenshot and nothing else: no element list, no locators, an empty A11y tab. The WebdriverIO service, Selenium (Node.js) and Nightwatch read the app's page source on every native action capture in trace mode (#569 for the latter two), so their native traces carry the element tree. Python is the one ✕ in the capability matrix's "Native app element list and accessibility tree per action (trace mode)" row.

Why it happens

instrumentation.py takes no per-action snapshot on a native app: the action-snapshot path is gated off for native sessions, so only the row screenshot is recorded and no page source is read.

What a fix needs

• Read GET /session/:id/source on each capturing command of a native session in trace mode, through the guarded executor so it renders no row and lands in no wait poll.
• Build the snapshot text and element locators from it. Core's fromPageSource (serializeMobileSnapshot plus generateAllElementLocators) does this in TypeScript; Python cannot import core, so either mirror it (with mirror-vectors.json cases) or have the backend build it, as it already does for other trace content Python cannot produce.
• Take the platform from the session capabilities (platformName, flat or under appium:options), as core's nativeAppOf does.
• Skip a webview's HTML page source, as Capture the native element tree per action in Selenium and Nightwatch #569 does.
• Time it as the pageSource probe under DEVTOOLS_CAPTURE_TIMING=1; it is the costly part of a native capture, most of it on iOS (Trace mode: measure where per-action capture time goes, per platform #516).
• Flip the Python cell in packages/conformance/src/features.ts, regenerate the README and docs-site tables.

Follow-up from #516 / #563.