feat(afana/zx-ir): phase normalization + strict range validation (closes #850)

[hiq-lab]

Summary

Adds spider-phase normalisation and a strict-mode validator to afana/src/zx_ir.rs, with integration tests in afana/tests/zx_phase_range.rs satisfying the issue's acceptance criteria.

New surface:

impl ZxGraph {
    pub fn normalize_phases(&mut self);
    pub fn validate_normalized(&self) -> Result<(), Vec<ZxValidationError>>;
}

pub enum ZxValidationError {
    // ...
    PhaseOutOfRange { node: NodeId, phase: f64 },
}

Why a separate strict variant

afana/src/lower.rs already produces un-normalised phases on purpose:

• Sdg → phase = -0.5
• Tdg → phase = -0.25
• Rx(theta), Rz(theta) → phase = theta / π for any real theta

Bolting a [0, 2) check onto the existing validate() would break the lowering pipeline on every parametric rotation. So the existing validate() keeps its current permissive behaviour (structural integrity + finite-phase check only), and the strict range check lives in a new validate_normalized() that callers opt into after running normalize_phases().

The intended workflow for emission paths:

graph.normalize_phases();      // semantics-preserving (rem_euclid(2.0))
graph.validate_normalized()?;  // strict gate before QASM
emit_qasm(&graph)?;

What normalize_phases() does

Phases in ZX-calculus are equivalent modulo 2π. The function calls phase.rem_euclid(2.0) on every finite phase, which produces a non-negative result in [0, 2). Non-finite phases (NaN/Inf) are left alone and continue to be flagged by the existing validate().

Tests

afana/tests/zx_phase_range.rs — integration tests per AC:

• validate_normalized_rejects_phase_negative_one_half ✓ AC QUASI-001: Ehrenfest CBOR Schema — base type definitions #1
• validate_normalized_rejects_phase_seven ✓ AC QUASI-002: HAL Contract Python bindings for quasi-agent #2
• normalize_phases_then_validate_succeeds — sanity check that the normalise → validate workflow accepts the same inputs after normalisation

afana/src/zx_ir.rs — 7 new inline unit tests covering edge cases (in-range acceptance, [0, 2) half-open at the upper bound, negative input, large positive input, NaN passthrough, exact 2.0 boundary, round-trip).

Test plan

• cargo test -p afana --test zx_phase_range — 3/3 pass
• cargo test -p afana --release — all 174+ existing tests still pass (no regressions in lower.rs, synthesis.rs, etc.)
• cargo build --workspace --release green

Follow-up worth considering

Calling normalize_phases() automatically inside lower.rs::lower_program() (or a new public lower_and_normalize() helper) would make the strict validation immediately usable everywhere. Not done in this PR — kept the change focused.