refactor: simplify Android hidden-content hinting pipeline

[thymikee]

Problem

The Android snapshot -i path currently computes hidden-content hints on the full snapshot tree and then copies hints onto the interactive tree via signature matching.

Current flow (interactive mode):

1. Parse XML tree once
2. Build full snapshot
3. Annotate native scroll hints (dumpsys)
4. Derive presentation hints on full tree
5. Build interactive snapshot
6. Copy hints full -> interactive using tight/loose signatures

This works, but it is hard to reason about and fragile because index/shape differ between full and interactive outputs.

Goal

Keep behavior unchanged for agents while reducing internal complexity and matching fragility.

Proposed direction

• Derive and attach hidden-content hints directly for interactive output.
• Use full-tree geometry only as context, not as the source object graph to copy from.
• Remove signature-based hint copy step if possible.

Non-goals

• No contract change for snapshot / snapshot -i output.
• No change to visible-first semantics.

Acceptance criteria

• snapshot -i keeps current hidden-content hint behavior on Android (Expensify/Settings style lists).
• Existing Android snapshot tests pass without reintroducing XML re-parse.
• Implementation no longer depends on tight/loose signature copy for full -> interactive hint transfer.