dtypes are inconsistent throughout rs.utils

[kmdalton]

We do not consistently use the same dtypes in functions within rs.utils. We should come up with a unified philosophy for how numpy dtypes are determined for returned values. I can think of at least three defensible possibilities:

• Use input dtypes to decide return dtypes
• Add a dtype=np.{float32, int32, ...} parameter to each function
• Always return np.float32 or np.int32 as applicable.

I lean toward the last option, because it meshes best with the mtz standard, and I think it will lead to fewer edge cases and gotchas.

[JBGreisman]

I agree with your take of prioritizing 32-bit dtypes where possible