Single render surface for memory retrieval

In ChatGPT, memory retrieval now renders a single full-content memory card instead of a truncated search-results card stacked on a duplicate detail card. Searching and listing memories stays fast and text-only; opening a specific memory is what shows the rich card.