feat(cmd): move worker setup out of init - #164

hand config reports the fleet's worker defaults and which of them are
still missing; hand config set validates one value and writes it
atomically, refusing a model or effort the configured harness takes no
launch flag for.

Model and effort are stored keyed by that harness (config/model.claude),
so switching harnesses re-asks instead of handing a worker an identifier
chosen for a different tool. An older home's unkeyed value is moved under
the harness it belonged to on init and update.

The bare command is the session hook, so it is where a supervising agent
learns the fleet is not configured yet. It now carries the same config
block hand config prints, and leads its help with one question per
missing value for the agent to put to the operator.

init now only bootstraps: it never reads stdin, asks nothing, and writes
no worker default, so it runs unchanged in a script, in CI, or with stdin
closed. What the fleet should dispatch is settled by the operator in the
first supervising session instead, because a value invented at bootstrap
time is indistinguishable from one the operator chose.

The --setup flag is gone rather than deprecated: nothing in the tool, the
tests or the session integration invoked it, and an unknown flag already
refuses with a usage document naming it.

…faults

The generated block now makes settling a missing worker default a step of
the workflow, and says who answers: the supervisor asks the operator in
conversation and persists the reply with hand config set. Answering its
own harness dialog would configure the fleet with the agent's guess.

The quick start built the tool from a checkout, which is the maintainers'
setup rather than an install, and pointed at hand init --setup for
configuration. It now installs a release binary, initializes a home that
asks nothing, and explains the worker defaults the first session asks for.
Building a checkout moves to CONTRIBUTING.md.

PR #165 made Codex model and effort capable before this rebase.
Configuration tests follow the merged harness contract instead of the
provisional matrix.

Refs #161