Fix frame-aware element extraction and click handling

[yifeizhou980429]

Changes in this PR

1. Add frame metadata to InteractiveElement
   Added a frame field to InteractiveElement
   Default is "main" for main-document elements

2. Make element extraction frame-aware
   Updated extract_elements() to iterate over all frames, not just the main page
   Each frame now runs element extraction separately
   Frame metadata is passed into the extraction script and stored on each element

3. Preserve iframe coordinates relative to the top page
   Added frame offset calculation so element coordinates from iframes are normalized relative to the top-level page
   This keeps element positions aligned with top-page screenshots / visual reasoning

4. Avoid cross-frame index collisions
   Reindexed extracted elements globally after aggregation
   This prevents duplicate element indices across main page and iframe contexts

5. Make click handling use frame-aware locators
   Updated click() to use frame-aware locator resolution
   If the element belongs to the main page, it uses page.locator(...)
   If the element belongs to an iframe, it resolves the correct frame and uses frame.locator(...)

6. Add safe fallback for missing frame metadata
   click() now safely defaults missing frame info to "main"
   This keeps the main-page interaction path robust even when frame metadata is absent

This fix makes browser interactions much more reliable for real-world pages that contain embedded frames.

Without this change:
iframe inputs may be matched but not clicked/focused correctly
typing may go to the wrong focus target
page-level keyboard events can happen instead of field input

With this change:
main-page and iframe inputs can both be selected and clicked correctly
focus is established on the intended target more reliably
typing is more likely to go into the correct field