refactor(cli, workspace): Extract lock acquisition

[JeanMertz]

The lock acquisition logic — polling at 500ms intervals, timing out after 30 seconds ($JP_LOCK_DURATION), and showing an interactive prompt on contention — was inline in query.rs. It is now extracted into cmd::lock and shared across all commands. As a result, config set, conversation edit, and conversation rm now poll and prompt on lock contention rather than failing immediately.

In jp_workspace, Workspace::lock_conversation now returns a LockResult enum instead of Option<ConversationLock>. The enum variant AlreadyLocked carries the ConversationHandle back so callers avoid an extra acquire_conversation round-trip. Two new methods — create_and_lock_conversation and its _with_id variant — atomically create and lock a conversation, removing the repetitive create-then-acquire-then-lock pattern across call sites.

ConversationLock now holds an Option<Arc<ConversationFileLock>> instead of a required Arc, so it works for in-memory workspaces without on-disk storage. The lock file placement falls back to workspace storage when user storage is absent, ensuring a flock can always be acquired.