Feature request: track incoming/outgoing tokens per session in Kimi vis, /status, and context status line

It would be very useful to see a per-session breakdown of token consumption, split into incoming (input/prompt/context) and outgoing (output/completion) tokens.

Where this information should be available:

1. Kimi vis — show input/output token counters alongside the existing visual context map.
2. /status command — add lines such as:

   tokens (session): 145.2k in / 38.7k out
   tokens (turn):    12.4k in / 3.1k out
   

3. Context status line — optionally display the counters next to the context usage indicator, e.g.:

   context: 66.8% (175.1k/262.1k) | tokens: 145.2k in / 38.7k out
   

Why this helps

• Users can quickly understand whether they are running out of context because of large file reads, tool outputs, or because the model is generating long responses.
• It makes it easier to optimize workflows and decide when to compact, restart, or switch to a model with a larger context window.

Suggested data to expose

• session_tokens_in / session_tokens_out — cumulative since the session started.
• turn_tokens_in / turn_tokens_out — for the current agent turn.
• Ideally also model/tokenizer-aware numbers when the provider reports them.

Start by tracing the existing context usage data and display paths for Kimi vis, the /status command, and the context status line. Then inspect how provider-reported token counts could map to session and turn totals. Done means the requested input/output counters are available on all three surfaces, with model-aware values used when reported.