Add `--model` / `-m` flag for headless (`-p`) mode

[vipulgupta2048]

Feature Description

Add a --model (and short -m) flag for headless / -p (print) mode so the model can be selected from non-interactive runs.

cmd -p "…" --model deepseek-v4-pro
cmd -p "…" -m deepseek-v4-pro

Accept the same model IDs the interactive /model picker shows.

Current state in v0.26.7:

• /model and Alt+P only work in TTY interactive sessions
• CMD_MODEL=deepseek-v4-pro cmd -p … is ignored — still returns the server-side default
• cmd config errors with Interactive mode requires a TTY
• Result: any script / CI / agent loop is locked to Claude Sonnet, regardless of prior interactive selection

Reproducer:

# interactive — set /model to DeepSeek V4 Pro, exit
cmd
> /model
> [select DeepSeek V4 Pro]
> /exit

# headless ignores the interactive selection
cmd -p "Reply with only your model identifier." --skip-onboarding --yolo
# → claude-3-5-sonnet-20241022

How other coding CLIs solve it (all support headless model selection):

codex exec --model gpt-5.5 "…"             # OpenAI Codex
kimi --print -m kimi-k2-thinking -p "…"    # Moonshot Kimi
gemini --yolo -m gemini-3.1-pro -p "…"     # Google Gemini
claude --print --model claude-opus-4-7 "…" # Anthropic Claude Code

Use Case

Wiring cmd into a multi-CLI factory automation loop (codex / kimi / gemini already integrated). The reason to add cmd to the loop is specifically DeepSeek V4 Pro — best open-source SWE-bench coder. Without headless model selection, cmd -p returns the same Claude Sonnet I already get from claude --print, so there is no reason to route through cmd at all. This defeats the value prop of having multiple model providers behind one CLI.

Additional Context

• Version: v0.26.7
• Flags tried that don't help: --skip-onboarding, --yolo
• Env vars tried: CMD_MODEL (ignored)
• cmd config is unusable in non-TTY contexts

How important is this to you?

Blocking adoption or production use