sklearn handles text labels differently than ml_dask on OneHotEncoding

[brian-methodical]

Describe the issue:
OneHotEncoding defaults to sets catagories_ to a numpy dtype='<U1' when explicitly given an array of text [["a", "b", "c"]] sklearn sets catagories_ dtype object.
Which is the correct behavior:

1. either cast catagories_ to dtype=object from OneHotEncoding on dask-ml when explicitly setting arrays (auto[1] is off)
2. or fix the test so it does not care about catagories_ dtype mismatch OR explicitly set dtype as argument in tests

[1] please note this issue is only with categories explicitly set and not when set to 'auto'.

When set to 'auto' dtype is numpy.float64 array, which is a third behavior. However we are matching what is done by sklearn there.

Minimal Complete Verifiable Example:

see test_basic_array

Anything else we need to know?:

This is illustrated in the failing test

Sklearn documentation found here shows dtype as object where dask-ml shows dtype as <U1 found here

Environment:

• Dask version: dask-ml-3.8 conda env dask 2023.1.1
• Python version: 3.8, 3.9, 3.10
• Operating System: ubuntu-latest
• Install method (conda, pip, source): conda