[BUG] dumps() emits extern calls without a semicolon, producing unparsable QASM #438

Environment

• pyqasm version: 1.2.0 (main @ 036ad4d)
• Python version: 3.12
• Operating system: macOS 15

What happened?

dumps() emits an extern function call without its terminating semicolon, so the generated
program is not valid OpenQASM and will not reparse.

import openqasm3, pyqasm

m = pyqasm.loads("""OPENQASM 3.0;
extern my_extern(int);
my_extern(3);
""")
m.unroll()
out = pyqasm.dumps(m)
print(out)
openqasm3.parse(out)

OPENQASM 3.0;
extern my_extern(int);
my_extern(3)

openqasm3.parser.QASM3ParsingError

With any statement following the call, the two run together on one line:

qubit[1] q;
my_extern(3)x q[0];      # line 5:12 no viable alternative at input 'my_extern(3)'

The cause is the node type, not the printer. The unrolled AST holds a bare FunctionCall
where a statement belongs:

>>> [type(s).__name__ for s in m.unrolled_ast.statements]
['ExternDeclaration', 'FunctionCall']

FunctionCall is an expression node, so openqasm3.printer renders it as an expression —
no semicolon, no line break. It should be wrapped in an ExpressionStatement.

Scope is narrow: only extern calls are affected. Subroutine calls and builtin calls both
round-trip correctly.

extern call        QASM3ParsingError   tail: ['qubit[1] q;', 'my_extern(3)x q[0];']
extern, no follow  QASM3ParsingError   tail: ['qubit[1] q;', 'my_extern(3)']
subroutine call    reparses OK         tail: ['x q[0];', 'x q[0];']
builtin call       reparses OK         tail: ['qubit[1] q;', 'x q[0];']

Suggestions (Optional)

Wrap the call in qasm3_ast.ExpressionStatement before it is appended to the output
statements — _handle_extern_function_cleanup (src/pyqasm/visitor.py:591) is where the
extern statement list is assembled.

Worth a round-trip assertion in the tests either way: openqasm3.parse(pyqasm.dumps(m)) over
the existing extern fixtures would have caught this, and would catch the same class of bug for
any other node that reaches the statement list unwrapped.

Found while reviewing #425, which touches the expression-statement path but does not cause or
fix this — it reproduces identically on main.