Proposal: Chat Flow context optimization

[leaft]

• 多个节点联合使用：选择多个previous node合并为一个“聚合”节点，fork 出一个新分支，然后继续对话，“聚合”节点应该需要主动触发一次上下文压缩，或许应该同时将“聚合“节点加入新分支的显式上下文缓存。

• 显式上下文缓存(branch 级别): 允许用户选择node加入显式上下文缓存，也许会有用，但似乎会让操作变得有些复杂。

• 顺带一提，chat flow chat 看起来很适合用于上下文缓存统计和可视化，在 research 或 prompt engineering 的场景。

• Multi-node Consolidation & Forking: Select multiple previous nodes to merge into a single "Aggregation" node, then fork a new branch to continue the conversation. This "Aggregation" node should trigger an explicit context compression (e.g., summarization or key-point extraction) and potentially be added to the explicit context cache of the new branch.

• Explicit Context Caching (Branch-level): Allow users to manually select specific nodes to be pinned into the explicit context cache. While potentially powerful, this might increase operational complexity for the user.

• Side Note: The Chat Flow interface seems inherently well-suited for context cache statistics and visualization, especially in Research or Prompt Engineering scenarios.