Fix Box.truncation discarding requested dims for array-backed boxes

[giodefelice-agents]

Closes #28, flagged for 🌙 Evening on memory#16's board. No verbatim human prompt — TODO.md quotes the issue instead, per the discopy#513/#514 precedent for self-contained bug+fix pairs.

The bug

Box.truncation's array branch built its result with Dim(2) ** len(dom/cod) unconditionally, ignoring the input_dims/output_dims its caller (Diagram.to_tensor) asks it to build against. That's correct for bit/qubit wires (always dimension 2) but silently wrong for any array-backed box with a mode/qmode wire above cutoff two — reproduced both symptoms from #28:

• a state (dom=Ty()) with a 3-entry array over one mode wire raised ValueError: cannot reshape array of size 3 into shape (2,)
• an endomorphism over one mode wire with a 3×3 array reported Dim(2) -> Dim(2) instead of Dim(3) -> Dim(3), then raised on eval()

determine_output_dimensions had the matching bug for the empty-dom case: it returned input_dims unchanged, which is [] for a state, so there was nothing to derive the codomain from.

The fix

Box.array_wire_dims and Box.array_truncation (public, no leading underscore, per STYLE.md's no-secrets rule — matches the pattern #29 is applying elsewhere) compute per-wire dimensions: bit wires stay fixed at 2, mode wires take the requested input_dims/output_dims. determine_output_dimensions falls back to the array's own shape when dom is empty, since that's the only case with no input to derive the codomain from. A size mismatch between the array and the requested dims now raises ValueError loudly instead of building a tensor.Box whose declared type doesn't match its data (the "worth deciding" question #28 raised — resolved in favour of the loud error).

Two per wire stays correct for bit/qubit, so nothing changes for the boxes that already used this path (zx.H, zx.SWAP — both override truncation directly and never reach the array branch anyway).

Tests

test/test_box_truncation.py — both #28 repro symptoms fixed and asserted, bit wires still fixed at 2 with no explicit dims, and the array/dims mismatch now raises.

Verification

• pflake8 optyx test/test_box_truncation.py: clean
• pylint optyx --fail-under=9: 9.55/10 (main: 9.54/10 per No secrets in channel.py and core/diagram.py, and make pylint report something #29)
• coverage run -m pytest: 1151 passed (1147 + 4 new)
• coverage report --fail-under=95: 95%

---

Generated by Claude Code