fix: Handle multi-parameter gate initialization correctly

[yurekami]

Summary

Fixes #232

This PR fixes the bug where multi-parameter gates (U2, U3, Rot, R, etc.) fail when passing 1D parameters like [0.1, 0.2].

The Problem

import torchquantum as tq

qdev = tq.QuantumDevice(n_wires=2, bsz=5, device="cpu")
tq.u2(qdev, [0], [0.1, 0.2])  # ❌ This was failing

The error occurred because gate_wrapper was transforming 1D params incorrectly:

• Before: [0.1, 0.2] (shape [2]) → [2, 1] via unsqueeze(-1)
• Expected: [0.1, 0.2] (shape [2]) → [[0.1, 0.2]] (shape [1, 2])

Multi-parameter gate matrix functions access params as params[:, 0] and params[:, 1], requiring shape [batch, n_params].

The Fix

Changed the dimension handling in gate_wrapper:

• 1D params: unsqueeze(0) instead of unsqueeze(-1) - adds batch dimension at front
• 0D params (scalar): unsqueeze(0).unsqueeze(0) - adds both batch and param dimensions

Affected Gates

These gates now work correctly with 1D input for single-batch usage:

• U2: tq.u2(qdev, [0], [phi, lam])
• U3: tq.u3(qdev, [0], [theta, phi, lam])
• Rot: tq.rot(qdev, [0], [phi, theta])
• R: tq.r(qdev, [0], [theta, phi])
• XXMinusYY, XXPlusYY: with [theta, beta]
• CU2, CU3, CRot: controlled versions

Backward Compatibility

The change maintains backward compatibility:

• 2D batched params [[0.1, 0.2], [0.3, 0.4]] still work as before
• Single-param gates with 1D input [0.1] → [[0.1]] still work correctly

Test Plan

• Run tq.u2(qdev, [0], [0.1, 0.2]) with bsz=1 - should work now
• Run tq.u3(qdev, [0], [0.1, 0.2, 0.3]) - should work
• Verify existing tests pass (no regression for single-param gates)
• Verify batched params [[0.1, 0.2], [0.3, 0.4]] still work

🤖 Generated with Claude Code