Store an Idle command's duration as time units instead of a rotation angle

[ciaranra]

Closes #727.

Problem

GateCommand in pecos-neo had only an angles field, so GateCommand::idle stored an idle gate's duration there by reinterpreting Angle64's u64 storage as a count of time units. The construction site said so outright. Every consumer that read angles therefore treated an idle duration as a rotation angle.

pecos-core already models this correctly and is unchanged here: GateType::Idle.angle_arity() is 0, and Gate::idle builds with empty angles and the duration in params. Only pecos-neo disagreed.

The visible consequence was in snap_command_queue, which snapped the duration to the nearest representable rotation angle: an idle of 23 time units became 0.

Change

pub struct GateCommand {
    pub gate_type: GateType,
    pub qubits: SmallVec<[QubitId; 4]>,
    pub payload: GatePayload,
}

pub enum GatePayload {
    Angles(SmallVec<[Angle64; 2]>),
    Duration(TimeUnits),
}

angles() returns &[] for a duration payload, so an Idle command genuinely has no angles and angle consumers need no special case for it. The duration is a TimeUnits throughout, exact across the full u64 range. new, with_angles, idle and get_idle_duration keep their signatures, so callers do not churn.

The consumers in queue_validation.rs change only from cmd.angles to cmd.angles(). Nothing there special-cases Idle, which is the point: the distinction now lives in the type rather than in each reader's memory.

pecos-core and the wire format are untouched. ByteMessage already derives angle count from gate type and writes non-angle parameters separately, and GateCommand has no serde implementation, so no encoding changes.

Incidental fix

sim_neo_bindings.rs contained a hand-rolled copy of the command-to-gate conversion that built a core Gate with params: GateParams::new() while copying the command's angles across. For an Idle command that wrote the duration into the angle slot of a gate whose duration belongs in params, so the experimental Python bridge was producing malformed Idle gates and dropping durations. It now calls the shared command_queue_to_gates.

Verification

• Release and debug: 5,133 tests and doctests each, zero failures, six pre-existing ignored.
• cargo clippy --locked --workspace --all-targets --all-features -- -D warnings: clean.
• cargo fmt --all -- --check: clean.
• Execution evidence: an idle of 23 snapped to 0 before this change and survives as 23 after. Durations of 0, 1, 23, 2^53, 2^53 + 1 and u64::MAX survive every snapping policy and report zero angles.
• Eight mutations were introduced and each was caught, covering duration rounding through f64, exposing a duration as an angle, dropping the duration during snapping, and dropping it on import, export and dispatch.

Converting a GateCommand into a core Gate still passes through TimeUnits::as_f64(), so durations above 2^53 lose precision at that boundary. That limit is pre-existing, follows from Gate.params being f64, and is entangled with the f64-based wire format, so it is deliberately left alone here.

Relationship to #701

#701 enforces gate angle arity. On its branch, pecos-neo special-cased Idle to report an angle arity of 1 so that the duration-in-angles storage would pass its own new validation, which is what made an idle duration visible to angle consumers as a rotation angle. With the storage corrected here, that special case is unnecessary.