Add wrappers for top unhandled commands: node, ssh, bun run, python, cat heredoc

Context

Running rtk discover on a real-world Claude Code session history (1277 sessions, 5255 Bash commands, last 30 days) reveals significant token-savings potential in commands RTK does not yet handle.

Top unhandled commands (by frequency)

Command	Count (30d)	Example
node	132	node dev-server.mjs
open	117	open "/path/to/file"
ssh	79	ssh -o StrictHostKeyChecking=no host 'cmd'
pkill	34	pkill -f "swift /tmp/find_videos"
ln	32	ln -s /src /dst
bun run	29	bun run dev
cat <<EOF	29	heredoc-style file creation
python / python3	39	python -m playwright install chromium
vercel	21	vercel --version

Estimated savings

rtk discover estimates ~50K additional tokens/month if these commands had RTK-equivalents (filtering output, suppressing progress bars, compacting JSON, etc.).

Suggested wrappers

• rtk node <script> — strip experimental warnings, dotenv noise, deprecation notices
• rtk ssh <host> <cmd> — same output filtering as rtk would apply locally to the remote command
• rtk bun run <task> — strip bundler progress + bun-internal noise
• rtk python / rtk python3 — filter pip install progress, virtualenv warnings
• rtk pkill / rtk ln / rtk open — even minimal compacting helps when called frequently

Why this matters

Even at the current ~1-2% adoption rate of RTK in mixed sessions, these top-9 unhandled commands account for roughly 5x the savings of all currently-handled commands combined for this user's profile. Users who do a lot of dev-server / SSH-driven workflows would benefit disproportionately.

Reproducing the data

Anyone with active RTK + Claude Code can reproduce:

rtk discover

The "TOP UNHANDLED COMMANDS" section is the input for prioritisation.