QFw: measure Braket programs in qubit order so every backend keys alike

[DougSO]

What

to_braket_qasm3 now writes every measurement as one block at the end of the program, c[k] = measure q[i_k] in ascending qubit order, into a register of exactly the measured qubits. The map it records still carries the circuit's own classical bits, and remap_counts rebuilds the Qiskit key from it as before, now filling every bit a qubit was measured into (the last measurement into a bit wins). A gate on a qubit after its measurement is refused before a task is created, since Braket takes terminal measurements only.

Why

The first permuted-measurement circuit run on a real Braket QPU came back wrong. The shim smoke's asymmetric circuit (X on q0, measured q2->c0, q0->c2, q1->c1) reads 100 on SV1 but read 001 on Rigetti Cepheus-1-108Q, for the same OpenQASM 3 program.

The raw task results explain it. Braket's backends do not agree on the order of a result's columns:

device	measuredQubits	row for the ideal shot
SV1	[0, 1, 2] (by qubit index)	100
Cepheus-1-108Q	[2, 1, 0] (by classical register)	001

Both are self-consistent, but amazon-braket-qdmi 1.2.0 (and main) builds its histogram keys by reversing each row as it comes, without reading measuredQubits (DeviceParser.cpp, parseMeasurementResults). So the key is "highest measured qubit leftmost" on SV1 and "highest classical bit leftmost" on Rigetti, and no fixed remap can serve both.

Making bit order, statement order and qubit order one and the same removes the ambiguity: every backend returns the same rows, the library's key reads highest measured qubit leftmost everywhere, and the existing remap is right. This also stays right if the library moves to classical-output semantics (munich-quantum-software/amazon-braket-qdmi-device#242): in the rank register, classical order and qubit order are the same order.

Verified

• tests/qiskit/test_braket_qasm3.py (12 passed, real Qiskit), tests/mock (631 passed, 1 skipped), ci-syntax clean.
• Live, through the shim QPM bound to each device, with this change installed:
  • SV1: X circuit {'1': 10}; asymmetric circuit {'100': 100}; Qiskit GHZ (1024 shots) via QFwBackend(provider="shim") {'000': 547, '111': 477}.
  • Rigetti Cepheus-1-108Q: asymmetric circuit, 100 shots, {'100': 77, '000': 19, '110': 3, '010': 1}, where the same circuit read {'001': 83, ...} before this change.

Notes

• Reuse released QDMI provider adapters with MQT Core 4.1 #147 rewrites the same function for the packaged adapters and will need a small rebase over this; the measurement block and the remap semantics here match what Reuse released QDMI provider adapters with MQT Core 4.1 #147 expects of remap_counts (one qubit may fill several bits, last write wins), so the two should compose.
• The docs section 8 and 13 describe the ordering and the terminal-measurement rule.