Enforce dtype for empty()

[certik]

LPython should give an error message:

 x: i32[4, 5, 2] = empty([4, 5, 2])
 y: f64[24, 100, 2, 5] = empty([24, 100, 2, 5])

Since in CPython the arrays would be both float (?) arrays of the default kind. Instead, it needs to enforce something like this:

 from numpy import int32, float64
 x: i32[4, 5, 2] = empty([4, 5, 2], dtype=int32)
 y: f64[24, 100, 2, 5] = empty([24, 100, 2, 5], dtype=float64)

[ubaidsk]

I think we can enforce it. In CPython, I think it is optional.

[certik]

What I mean by that is that if dtype is not specified, then it will use the same dtype as CPython (I think it's float64).

Consequently, this x: i32[4, 5, 2] = empty([4, 5, 2]) must fail, because the RHS is f64 and the LHS is i32.

The issue is that if we don't do that, then the above code has different types in CPython and LPython, so LPython behaves differently, which is breaching our contract that if LPython compiles the code (and runs), then CPython will behave 100% the same.

[certik]

I am assigning this to @czgdp1807, since he implemented #2118 for i8, so it just needs to be extended for all dtypes.