Bugfix (ensure return datatype matches arguments)

This commit fixes a bug where calling this function with single-
precision arguments would produce double-precision results.

This bug is unlikely to have caused any issues, however, because it is
essentially unreachable. This code is only executed when use_mex=true
but the corresponding MEX file is unavailable, but it is difficult to
enter that state because you can't set use_mex=true unless the MEX
files are present.

@MoDT/bandPosSolve.m

@@ -19,6 +19,7 @@
 [p,N] = size(A_bands);
 
 % MATLAB doesn't support single-precision sparse yet (as of R2017a)
+was_single = isa(A_bands,'single') || isa(b,'single');
 A_bands = double(A_bands); b = double(b);
 
 % Call sparse() to construct A
@@ -32,7 +33,7 @@
 x = A \ b;
 
 % Cast to single if A or b was single
-if isa(A_bands,'single') || isa(b,'single')
+if was_single
     x = single(x);
 end