[WIP] Broadcast and Merge library nodes, reachable from NumPy

[ThrudPrimrose]

Adds two standard library nodes -- Broadcast (Fortran SPREAD with an integer dim, the right-aligned NumPy rule with dim=None) and MergeLibraryNode (the per-element select: Fortran MERGE, np.where, element-wise if-then-else) -- both describing every operand by the same broadcasting rule instead of a special case. numpy.where hands the node the case it expresses exactly (three arrays, no cast, a condition that broadcasts into the result) and keeps the inlined tasklet otherwise, and numpy.broadcast_to is new.

@dace.program
def k(a: dace.float64[N, 1], b: dace.float64[N, 4], c: dace.bool_[N, 4]):
    return np.where(c, a, b)     # (N, 1) operand read at column 0 for every column