feat(trusty-mpm): add n [name] picker command for launching a named session

[mac-duetto]

Closes #4965

Adds an n command to the interactive session picker (bare tm / tm ls) so a new session can be launched with a meaningful name instead of the daemon-derived tm-<repo>-NN.

Grammar

Input	Result
n / N (exact)	Launch new, auto-named
n <name> / N <name> (separator required)	Launch new, named
new, no, nn, n1, nauth	Unrecognised — unchanged

The integer "launch new session" entry is unchanged. This is purely additive.

Two commits

c8ccb317 — the feature as originally specified in #4965.

7fdbf4d7 — fixes from an adversarial review that returned WARN. The first commit used strip_prefix(['n','N']), which consumed any n-prefixed input and returned a decision that unconditionally clones a repo, creates a worktree, starts tmux, and attaches — with no confirm step. The menu line literally reads launch new session, so an operator typing new would get a real session named tm-ew-01. The delete confirm is spelled [y/N], so n already means "no" inside this tool.

The d<N> / r<N> <name> no-space precedent is only safe because their remainder must resolve to a live slot — d9 on a 3-session list is Unrecognised. n had no such gate. It is now an exact-match token plus a required separator.

Sanitization moved CLI-side

Previously the raw string was forwarded and the daemon sanitized it. Two bad outcomes:

• n feature/auth and n hotfix/auth both produced tm-auth-NN — the daemon's resolve_session_name routes the hint through leaf_slug_from_dir, whose Path::file_name() discards everything before the last /. That defeats the point of the feature.
• n !!! sanitized to empty and fell back to tm-local-NN, a name in a leaf namespace shared with every other project on the daemon and less identifiable than the default.

The name is now kebab-cased in the picker before it is sent. Rather than exposing the private sanitize_slug (which would also mean exposing MAX_FOLDER_LEN, inviting drift), this adds one public item — leaf_slug_from_hint in trusty-common — and refactors leaf_slug_from_dir to delegate to it. No second kebab-caser. leaf_slug_from_hint_is_a_fixed_point_for_its_own_output proves the daemon's own pass is a no-op rather than asserting it.

A name that sanitizes to empty is now rejected with a message instead of silently spawning. When the sanitized name differs from what was typed, one line is printed before launch: tm: naming it 'my-auth-fix' (from 'My Auth Fix!').

Other changes

• The command legend moved out of session_picker.rs into session_picker_render.rs as command_legend(Option<u32>) -> Vec<String>. Inline, the file hit 515 SLOC against the repo's 500 cap and check_line_cap.sh refused it. This also made the column alignment and the tm-<name>-NN wording assertable — 4 tests.
• The n legend line is self-documenting (launch new session as tm-<name>-NN), since the adjacent r<N> <name> takes a verbatim full session name while n <name> takes a leaf.
• launch_new_session_and_attach's name_hint plumbing had zero coverage. Two wire tests now assert the key is present on Some and absent (not null) on None.

Test ladder — rung 4

trusty-common gained public API, so this is a cross-crate change.

cargo fmt --check                                        → 0
cargo clippy -p trusty-mpm --all-targets -- -D warnings   → 0
cargo clippy -p trusty-common --features session-naming   → 0
scripts/check_line_cap.sh                                 → 0
cargo test -p trusty-common  → 340 passed; 0 failed; 6 ignored
cargo test (lib)             → 4636 passed; 0 failed; 7 ignored
cargo test (bin tm)          → 1491 passed; 0 failed; 1 ignored
cargo test (bin trusty-mpm)  → 1491 passed; 0 failed; 1 ignored
cargo check --workspace (excl. GUI crates) → 0

Pre-existing failures on this base — not from this change

Each was verified against the stashed base before being set aside. Nothing was #[ignore]d or narrowed to make a gate green.

• ensure_managed_config_dir_emits_the_frozen_skill_warning — fails ~1 run in 4, passes in isolation. #[serial] only serializes against other #[serial] tests, and installing a subscriber perturbs the process-global interest cache. Failed 1-of-4 on base.
• compose_session_instructions_display_matches_live_prompt[_with_override] — tm and trusty-mpm are duplicate bin targets over the same main.rs, so cargo test -p trusty-mpm runs the test twice concurrently against the same on-disk project state. Failed 1-of-3 on base; green when the targets run separately.
• cargo check --workspace — trusty-code-gui and trusty-mpm-gui panic in tauri::generate_context! because ui/dist is not built in this worktree. Identical exit 101 on base.

CI may be flaky on this PR for those reasons, unrelated to the change.

🤖🤖🤖 Generated with trusty-mpm — https://github.com/bobmatnyc/trusty-tools