fix(web-ui): verify message source before applying widget updates (light-tested)

[1688mengdie]

Summary

The generative widget iframe shell listened for message events without
checking where they came from. Any window able to postMessage into the
host page could forge a bitfun-widget:update event and have arbitrary
attacker-controlled HTML applied inside the widget document via
setContent.

This PR rejects every message whose source is not the direct parent
window before any data processing, mirroring the source check already
applied to the iframe-to-parent direction. An anti-injection test
asserts the gate sits ahead of setContent in the widget shell.

Fixes #2603

Type and Areas

Type: security fix

Areas: Web UI

Verification

• pnpm --dir src/web-ui run type-check — passed (0 errors).
• pnpm --dir src/web-ui run lint — passed (0 errors).
• vitest run GenerativeWidgetFrame.test.tsx — 3 passed (including the new anti-injection case; the two pre-existing cases are untouched).

Reviewer Notes

• The guard is a one-line source check at the top of the message listener, covering the data path and the clear-selection branch alike; behavior for legitimate parent-originated updates is unchanged.
• No new dependencies and no API changes.
• Rollback is a revert of the one-line guard plus the appended test case.

Checklist

• This PR is focused and does not include secrets, temporary prompts, generated scratch files, or unrelated artifacts.
• Relevant verification is recorded above, or skipped checks are explained.
• User-facing strings, docs, and locales are updated where applicable. (Not applicable: no user-facing change.)

This is an AI-assisted change.