Toggle Taste learnings displays

先定位負責呈現 TASTE Learned、Reinforced 和 Moved 訊息並處理 taste.md 更新的 CLI 程式碼；該 issue 未指定特定的原始檔案或測試。新增一個由使用者控制的切換開關，在抑制這些顯示的同時保留對 taste.md 的學習，並確認靜默更新仍會寫入 taste.md，而正常輸出不受影響。

Summary

I've had extreme amounts of output from taste.md learnings, so much so that sometimes the output from taste.md is much more that the actually needed output. In such cases it would be handy to have a toggle option to keep toggle taste learning display.

Here is an eg
output is interleaved with taste info + 2 blocks of taste op for each block of actual info

❯ ok then give me a better way 
  explain approach 1 to me     

✻ Thought for 1 second [ctrl+o to expand]

⠶ Approach 1 — sizeof works because the compiler knows the array size at compile time in
  the same scope where it was declared:

  char *a[] = {"hello", "world"};
    size_t n = sizeof(a) / sizeof(a[0]);   // 2

  sizeof(a) gives the total bytes — 2 pointers × 8 bytes = 16 on 64-bit. sizeof(a[0])
  gives one pointer = 8 bytes. 16/8 = 2. It's computed at compile time, no runtime cost,
   no sentinel needed.

 TASTE  Reinforced
 └ Asks conceptual "why" questions to understand the underlying logic/reasoning behind
   a suggested fix before applying it, rather than blindly following instructions 
   [Confidence 70% → 80%]
   taste.md

 TASTE  Reinforced
 └ Asks conceptual "why" questions to understand the underlying logic/reasoning behind
   a suggested fix before applying it, rather than blindly following instructions 
   [Confidence 70% → 80%]
   --prefers-the-assistant-to-first-read-and-understand-the-problem-spec-(e.g.,-lab-pdf
   )-and-existing-code-progress,-then-wait-for-explicit-instructions-before-making-chan
   ges.-confidence:-0.7/taste.md

  But it only works when a is the actual array. Pass it to a function, it decays to
  char** and sizeof(a) becomes just the pointer size (8), so the trick fails.

  The better way for runtime use: pass the size along with the array.

  void f(char **a, size_t count);

  That's what every real C API does (argc/argv style). It's simple, explicit, never
  reads out of bounds, and you know the count the moment you built the array — no
  guessing, no sentinel, no UB.

 TASTE  Moved
 └ - Prefers the assistant to first read and understand the problem spec (e.g., lab
   PDF) and existing code progress, then wait for explicit instructions before making
   changes. Confidence: 0.7
   --prefers-the-assistant-to-first-read-and-understand-the-problem-spec-(e.g.,-lab-pdf
   )-and-existing-code-progress,-then-wait-for-explicit-instructions-before-making-chan
   ges.-confidence:-0.7/taste.md

 TASTE  Moved
 └ - Prefers the assistant to first read and understand the problem spec (e.g., lab
   PDF) and existing code progress, then wait for explicit instructions before making
   changes. Confidence: 0.7
   --prefers-the-assistant-to-first-read-and-understand-the-problem-spec-(e.g.,-lab-pdf
   )-and-existing-code-progress,-then-wait-for-explicit-instructions-before-making-chan
   ges.-confidence:-0.7/taste.md

 ✻ Worked for 7s

So just to clarify I want taste learning, but I do not want to annoyed by its "learning" each and every time

Expected Behavior

Silent updates to taste.md

Actual Behavior

I asked it to stop showing me these outputs and this is the response.

 TASTE  Learned
 └ Does not want taste-learning updates/pop-ups surfaced or referenced during the
   conversation; still wants the assistant to keep learning their taste silently in the
   background [Confidence 95%]
   taste.md

Steps to reproduce the issue

Do anything

Command Code Version

1.39.2

Operating System

Linux

Terminal/IDE

VSCode

Shell

zsh

Session file (optional)

No response

Fix prompt (optional)

No response

Additional context

No response