Preserve StatePreparation width in inverse

[TSS99]

Summary

Fix #16430.

StatePreparation.inverse() rebuilt the inverse gate from the stored integer
value alone, so an explicit num_qubits given at construction was lost. For
integer inputs the constructor then re-inferred the minimal width from the
value, e.g. StatePreparation(1, num_qubits=2).inverse() collapsed to a
1-qubit gate.

The fix forwards the original width when rebuilding the inverse, but only for
integer-backed instances — num_qubits is invalid for label and vector inputs,
so those paths are unchanged.

Tests

Added a regression that appends StatePreparation(1, num_qubits=2) and its
inverse on the same 2-qubit circuit and checks the result is the identity.

python -m unittest test.python.circuit.library.test_state_preparation -v

AI/LLM disclosure

• I didn't use LLM tooling, or only used it privately.
• I used the following tool to understand this PR: Claude and ChatGPT

[Cryoris]

Let's give #16431 a chance first, that was opened before and has the exact same fix.