Report the session title from the turn, not from its end

[omartelo]

Client side of omartelo/lich#418, which moves the contract's event mapping. The payload does not change, so no fixture does either.

Why

Claude Code fires its title call in parallel with the turn's own first model call — the ai-title is on disk about three seconds after the prompt. Reporting only on Stop meant a first turn that runs for ten minutes showed the card's placeholder name for all ten, and an interrupted turn never got a name at all, with the real title sitting there the whole time.

Codex and Antigravity are worse served still: their title is the user's own first message, which exists before the turn does.

What changed

harness	before	now
Claude Code	Stop	PostToolUse + Stop
Codex	Stop	PostToolUse + Stop
Antigravity	Stop	PreInvocation + Stop
opencode, omp	unchanged — both already report during the turn
Crush, Cursor	unchanged — neither has a title to report

Antigravity gets PreInvocation because its PostToolUse is matched to write tools only, and PreInvocation is its earliest event either way.

The once argument

The in-turn registrations pass once, and that is the whole difference: the script latches on the first title lich accepts and stops looking. Finding a title means scanning a transcript that grows all turn — 324ms on a 168MB one — and the in-turn hook fires per tool call, so without the latch a fifty-tool turn pays that fifty times to learn what it knew after the first.

• The latch is an empty $TMPDIR/lich-title-$LICH_SESSION_ID; its existence is the signal, and it is checked before stdin is even read.
• curl gains -f, so a refusal counts as one and does not latch — a title lich did not take is worth looking for again next call, rather than waiting for Stop.
• Stop runs unlatched and keeps scanning. That is the path a later retitle arrives on.

Tests

runHook now gives every hook a scratch $TMPDIR of its own — a hook keeps state there and every test here shares one session id, so without it the latch leaked between tests and into the developer's /tmp. A test that means to exercise what carries across calls passes the same directory to both.

Three new cases: once reports one title and then stops; the unlatched form reports every time and carries a retitle; a refused report does not latch.

• node --test tests/*.test.mjs — 194 pass (was 191)
• Mutation-checked: removing the latch fails session-title reported once per session stops scanning
• sh -n clean on every hook script
• Fixtures untouched, so the drift job stays green