`pow(k) @` modifier rejects non-integer exponents

[TheGupta2012]

Limitation

The power modifier is defined as: "The modifier
pow(k) @ replaces its gate argument U by its kth power U^k for some positive integer
or floating point number k (not necessarily constant)."

pyqasm requires k to be an integer and rejects everything else.

Example QASM failure

OPENQASM 3.0;
include "stdgates.inc";
qubit[1] q;
pow(0.5) @ x q[0];

ValidationError: Power modifier argument must be an integer in gate operation QuantumGate(...)

Integer exponents, including negative ones, work today:

pow(2) @ x q[0];    // OK
pow(-1) @ x q[0];   // OK

so the restriction is specifically on the fractional case — which is the interesting one, since
pow(0.5) @ x is the square root of X.

Change Requested

1. Accept a floating-point k in pow(k) @.
2. For a fractional k, the repeat-the-gate expansion used for integers does not apply. The
   unrolled result must be computed from the gate's matrix as U^k via matrix power — the
   principal branch, matching the spec's definition.
3. Where a fractional power of a gate has a known closed form in the standard library, prefer
   emitting that named gate over a synthesised U(θ, φ, λ). For example pow(0.5) @ x is
   sx.
4. If a fractional power cannot be realised for a given gate, raise a ValidationError naming
   the gate and the exponent — not the current blanket "must be an integer".

Implementation Details

• The rejection is at src/pyqasm/visitor.py:1627. The surrounding block already handles
  negative integers by inverting and repeating, so the integer path is a useful reference for
  where the fractional branch belongs.
• src/pyqasm/linalg.py and the Cython src/pyqasm/accelerate/linalg.pyx already provide the
  matrix machinery used by the decomposer; a matrix power via eigendecomposition fits there
  rather than in the visitor.
• The single-qubit result can be routed through the existing ZYZ / U(θ, φ, λ) decomposition
  in src/pyqasm/decomposer.py, so the fractional case reduces to "compute the matrix, then
  decompose" and reuses tested code.
• Decide the multi-qubit policy explicitly. A fractional power of a two-qubit gate is well
  defined mathematically but expensive to synthesise; rejecting it with a clear message is a
  reasonable first increment, provided the message says so.
• The spec's "not necessarily constant" clause implies a runtime-valued exponent. That cannot
  be resolved during unrolling and is out of scope here — reject it with a specific message.
• Tests: tests/qasm3/test_gate.py — pow(0.5) @ x against the sx matrix, pow(0.25) @ x,
  pow(1.5) @ z, a fractional power under ctrl @, and a clear error for the unsupported
  multi-qubit case.