fix(hooks): keep the hook protocol channel free of the setup nag

[rhishi99]

Problem

rtk hook and rtk rewrite are the hook. The agent that invokes them parses their stdout and stderr as a protocol channel, so an extra line there is corruption, not a diagnostic.

maybe_warn() was allowed to fire for both. On a machine that has Claude Code installed but not the rtk hook, that puts

[rtk] /!\ No hook installed - run `rtk init -g` for automatic token savings

into the protocol stream once a day. Gain was already excluded for exactly this reason; Hook and Rewrite were missed.

This is not theoretical

On such a machine, ten tests in tests/copilot_selfheal_test.rs fail today — every one on the same assertion:

stderr must stay silent (protocol safety) for payload {"tool_name":"Bash",...},
got: [rtk] /!\ No hook installed - run `rtk init -g` ...

CI never sees it because a fresh runner has no ~/.claude at all, so status() returns Ok and the warning never fires. The tests are correct; the nag was leaking into a channel it must never touch.

Fix

Add Commands::Hook and Commands::Rewrite to the same exclusion list Commands::Gain already sits in.

Testing

No new test — copilot_selfheal_test.rs already pins this contract with 12 assertions on stderr silence; they were simply unreachable in CI. Before/after on an affected machine:

before: test result: FAILED. 2 passed; 10 failed
after:  test result: ok.    12 passed;  0 failed

Full suite, cargo fmt --check, and cargo clippy --all-targets clean on x86_64-pc-windows-gnu.

🤖 Generated with Claude Code

[CLAassistant]

All committers have signed the CLA.