Feature Request: Add show_tps_meter config and /tps command for TPS display

[leonidlouis]

Summary

Add ability to display tokens-per-second (TPS) meter in the status bar during LLM streaming via config option and slash command.

PR is #1759

Motivation

Currently, users have no visibility into how fast tokens are being streamed. This feature helps users:

• Monitor streaming performance in real-time
• Understand when the model is actively generating vs idle
• Compare speeds across different models/providers
• It's cool to see. - inspired by This Extension of Opencode

Proposed Solution

Config Option

In ~/.kimi/config.toml:

show_tps_meter = true  # Default: false

Slash Command

In kimi shell:

/tps on    # Enable TPS meter
/tps off   # Disable TPS meter
/tps       # Show current status

Behavior

Scenario	TPS Display
show_tps_meter = false (default)	Hidden
show_tps_meter = true, streaming	Shows "· 85.3 tok/s" in status bar
show_tps_meter = true, not streaming	Hidden (no tokens flowing)
/tps on during session	Enables immediately, persists to config
/tps off during session	Disables immediately, persists to config

Implementation Approach

The implementation mirrors the existing theme pattern for global UI state:

1. UI State Module (ui/tps_meter.py): Global getter/setter for the toggle state
2. Config Field (config.py): show_tps_meter: bool = Field(default=False)
3. Slash Command (ui/shell/slash.py): /tps on|off with persistence
4. TPS Tracking (soul/kimisoul.py): Rolling 3-second window with CJK-aware token estimation
5. Display Integration (visualize.py, prompt.py): Conditional display in status bar and bottom toolbar

Technical Details

• Rolling Window: 3-second sliding window for TPS calculation (configurable)
• Token Estimation: CJK-aware heuristic (1.5 tokens per CJK char, 0.25 per ASCII char)
• Performance: Zero overhead when disabled (short-circuit evaluation)
• Pattern: Mirrors theme.py for consistency with existing codebase conventions

Checklist

• CLI slash command (/tps)
• Config option (show_tps_meter)
• Proper precedence logic (runtime toggle > config > default)
• Session persistence (config saved on toggle)
• Unit tests (19 tests covering calculation, state, slash command, display)
• Type checking passes (pyright)
• Linting passes (ruff)