Request support for codewhale/deepseek-tui

## description
deepseek-tuihas been renamed and rebranded as CodeWhale​ — a terminal coding agent that supports arbitrary models (DeepSeek, GLM, Kimi, MiniMax, OpenRouter, Arcee, local vLLM/SGLang/Ollama, etc.) with a Fin auto-routing layer. Config lives under ~/.codewhale/(e.g. ~/.codewhale/mcp.json, provider credentials, skills dir, etc.).

abtop currently supports Claude Code​ and Codex CLI​ as first-class agents (process scan + config-dir discovery + session/transcript parsing). It would be great to add CodeWhale as a third supported agent so users running mixed stacks (Claude + Codex + CodeWhale) get a unified TUI view.

Proposed scope (mirroring the existing Claude/Codex collectors):
- Discover running codewhaleprocesses (the TUI runtime + the codewhalescheduler binary).
- Read ~/.codewhale/ to surface active sessions, token counts, context-window usage, and estimated cost (CodeWhale supports 25+ providers, so per-provider pricing lookup will be needed — similar to how Claude Code maps model → price).
- Rate-limit / quota panel: token throughput + remaining context can still be gauged from session transcripts.

## Use case
I run several agents concurrently: Claude Code for one repo, Codex for another, and CodeWhale (DeepSeek / Kimi via OpenRouter) for long-running refactors where I want cheaper tokens. Right now abtop shows Claude + Codex beautifully, but the CodeWhale panes are invisible — I have to ps/ check ~/.codewhalemanually to see which session is burning context or forgot to close a dev-server port. Adding CodeWhale would make abtop a true "any-local-AI-agent" top for my workflow.

## Additional context
CodeWhale is the successor to deepseek-tui(npm: npm install -g codewhale), maintained by Hmbown, codewhale 0.8.64 (9d2e6a89f586) at time of writing.

Happy to help test / iterate if a branch goes up.

Start by tracing the existing Claude Code and Codex collectors for process scanning, config-directory discovery, and session/transcript parsing. Then inspect how their model pricing and TUI panels expose tokens, context usage, costs, and rate limits. Done means CodeWhale processes and ~/.codewhale/ sessions appear in the unified view with the proposed usage information.

Written by the indexing model from the issue text.