[TUI Improvement] Group taste feedback messages in the TUI

Feature Description

I've checked out that, whenever you say / set up a ton of preferences usually taste-1, creates a ton of feedback texts, that push the tui text up (if you are watching what your model made, etc.., it automatically pushes everything up), in the image there's only 2, but for example, there's some times, that a ton of tastes feedbacks appear and its kinda annoying to deal with.

So i was trying out if i could solve it via a mod implementation, and yeah couldn't make it, as on the 2nd image appears the thing is to have all tastes stored inline, for example a chevron or submenu such as the one i tried to make with all taste implementations down there, instead of:

TASTE Learned
└ Prefers TASTE Learned notifications grouped into a single collapsible chevron when 2+ appear (English
header, standard format inside, open to adapted compact format) and suppress/hide individual notices to
avoid duplication/spam [Confidence 92%]
taste.md

Another one that says

TASTE Learned
└ Prefers TASTE Learned notifications grouped into a single collapsible chevron when 2+ appear (English
header, standard format inside, open to adapted compact format) and suppress/hide individual notices to
avoid duplication/spam [Confidence 92%]
taste.md

etc...

Something like what i tried to make

◼ TASTE Learned ▸ (down clickable chevron with all tastes)
└ Prefers TASTE Learned notifications grouped into a single collapsible chevron when 2+ appear (English)
└ more taste changes...
└ more taste changes...

Use Case

• Keeps the learning, kills the disruption — you still get the personalized taste profile, but the transcript stops jumping while you watch output.

• Reduces cognitive load — instead of a wall of repeated feedback lines, one grouped/expandable notice tells you learning happened without stealing focus (with a chevron to check out which other taste implementations have been made), like a minimization tool.

• Better for review flow — you can watch long generations end-to-end without the viewport scrolling away, which is exactly the workflow taste is supposed to support

Additional Context

No response

How important is this to you?

None

首先定位 TASTE Learned 反馈消息的 TUI 渲染路径，并跟踪各条通知是如何添加到转录中的。为两条或更多消息实现带有可展开指示器的分组，同时保留已学习的偏好，并防止单条通知重复；验证长时间生成不再导致重复的反馈行移动 viewport。