CLI-119: lane next has three unhandled failure modes (start race, create-then-type stall, slug collision) #2576

Found by the independent review of CLI-119-lane-next (FAIL at ff0b161, PR #2573). Three minors the fix PR dispositions as tickets, not fixes.

1. Find -> Create race (cli/internal/lane/handoff.go). Two sessions started at once both see no worktree for the branch and both call Create. AC3 ("or from two sessions") is proven only sequentially. Fix options: serialise on the lease file / a lock under the git common dir, or treat a Create conflict as "already started" and re-Find. Add a parallel-run test. The atomic claim of CLI-128 (CLI-128: atomic lane claim on a git ref #2551) may subsume this.
2. Create succeeds, Type fails. The branch is left with an open workspace and no agent; every later run answers "a workspace already shows ; no second agent started", so the queue stalls until a human closes it. Fix options: on a Type failure close the workspace this run opened (Backend has no remove today, by design), or let a re-run type again when the pane shows no agent.
3. Slug collisions. Two issues whose titles slug alike, or a title with no [a-z0-9] word (slug "lane"), share one branch, so the second is never started: Find sees the first's worktree. Fix options: add the issue number to the branch name, or detect the collision in Next.

Root cause common to all three: the hand-off decides from a read (Find) and acts later, with the branch name as the only identity.