Can we brainstorm a namespace for array creation/inspection utils? Maybe as part of np.lib.<submodule>, maybe directly np.<submodule>?

Functions I might dump there are:

• byte_bounds (apparently some use it, and it's an OK function, but not really a general audience one, unlike may_share_memory which I think makes sense at least for teaching).
• np.core._multiarray_umath._discover_array_parameters: Finds the shape and dtype the same way as np.asarray(), but doesn't create the full array (although it does create some temporaries). Seems to be useful occasionally, and may also be useful to add additional functionality that we do not want in np.asarray().
• prepare_arrays(*arrays, *, kwarg=None) may be useful. For NEP 50, maybe asarray_or_pyscalar makes just as much sense (one can trivially write that admittedly). @mdhaber liked the idea of having a function that converts multiple arrays to their result type (which could be an option there). I like the idea of clumping asarray calls; opitonally including result_type seems convenient (even if it may not always be ideal as it assumes simple promotion is correct, it is effectively what is often used anyway).
• normalize_axis_index
• Probably we can think of more functions to add there or move?

Maybe utils is already a decent name and it just needs to lose some of the weirder functions?