[FEATURE] Decompose p, phaseshift and u1 as rz instead of h; rx; h #469

Feature Description

unroll decomposes p, phaseshift and u1 into three gates, h; rx(θ); h. A single rz(θ) gives the same gate up to a global phase. The longer form then grows during rebase, because each h expands again:

Reproduction on main at a742796, Python 3.12:

import pyqasm
from pyqasm.elements import BasisSet

src = 'OPENQASM 3.0;\ninclude "stdgates.inc";\nqubit[1] q;\np(pi/4) q[0];\n'
m = pyqasm.loads(src)
m.rebase(BasisSet.CLIFFORD_T)
print(pyqasm.dumps(m))  # h q[0]; h q[0]; t q[0]; h q[0]; h q[0];

OpenQASM 2 programs that use qelib1.inc unroll p and u1 the same way.

The decomposition is phaseshift_gate. ONE_QUBIT_ROTATION_MAP maps u1, U1, phaseshift and p to it. The csx and rccx decompositions also call it seven times in total.

A related point: #468 adds a p rule to rebase. Unroll rewrites a p gate written by the user before rebase sees it, so that rule only applies to the p gates that pyqasm generates itself.

Done when:

• A single p(θ), phaseshift(θ) or u1(θ) unrolls to one gate.
• Under CLIFFORD_T, p(pi/4) rebases to t and p(pi/2) rebases to s.
• The unitary of each changed decomposition, including csx and rccx, still matches the original up to a global phase.

Implementation (Optional)

One option is to make phaseshift_gate emit rz(θ). gate_decompositions.md documents the current form and needs the same update. rz(θ) and p(θ) differ only by a global phase. That should be safe here, because ctrl @ p resolves to cp before this decomposition runs. This reasoning is not verified against the full test suite.