docs: update Video Chunk Sampling section to use patch_positions

[Copilot]

Updates the "Video Chunk Sampling" section in docs/model_card.md to reflect the actual API, which uses patch_positions instead of visible_indices.

Changes

• Updated flow diagram to reference patch_positions for temporal mapping
• Renamed mechanism section from visible_indices to patch_positions
• Rewrote code example to show building [batch_size, seq_len, 3] tensor with [t, h, w] coordinates

Example

# Build patch_positions: [batch_size, seq_len, 3] with [t, h, w] for each patch
t_positions = frame_pos.unsqueeze(-1).expand(-1, frame_tokens).reshape(-1)
h_positions = h_ids.unsqueeze(0).expand(num_frames, -1).reshape(-1)
w_positions = w_ids.unsqueeze(0).expand(num_frames, -1).reshape(-1)

patch_positions = torch.stack([t_positions, h_positions, w_positions], dim=-1)
patch_positions = patch_positions.unsqueeze(0)  # [1, 4096, 3]

Original prompt

Please update the "Video Chunk Sampling" section in docs/model_card.md first, changing the input method from visible_indices to patch_positions.

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.