fix(ode): validate ExponentialMap::integrate_attitude's timestep and … - #307

…rate

Fixes kmolan#302. attitude_step and attitude_step_with_angular_acceleration fed a
timestep and angular rate straight into SO3::exp with no validation: a
non-finite rate produced a quaternion of NaNs that never recovered, and a
non-positive timestep integrated backwards (or not at all) without comment.

attitude_step and attitude_step_with_angular_acceleration stay infallible,
since they sit on RigidBody::stepped's documented panic-free per-tick path
(and estimation::attitude_correction's per-tick path) - making them fallible
would force that hot loop to become fallible too. Their docs now carry an
explicit "Behavior" note describing the NaN policy instead of leaving it
silent.

integrate_attitude, the multi-step driver (not used on any per-tick hot
path, and already the same shape as the fallible Rk45::solve), now validates
dt up front and the rate returned by its angular_rate_at callback on every
step, returning the new IntegrateError::NonPositiveTimestep or the existing
IntegrateError::NonFinite instead of silently producing garbage.

- error.rs: fix indentation of the NonPositiveTimestep Display arm
- tests/suite/ode/exponential_map.rs: fix import order and split
  chained .unwrap() calls per rustfmt
  - tutorials/ode-integrators.md: add .unwrap() to the now-fallible
    integrate_attitude doctest call, and update the "Errors" paragraph
      to reflect that attitude_step/attitude_step_with_angular_acceleration
        stay infallible while integrate_attitude now returns IntegrateError

Result is already #[must_use]; the extra attribute trips clippy::double_must_use under -D warnings.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>