Add support for numpy arrays with unsigned integer types

Currently it is impossible to create a datatable frame from numpy arrays with unsigned integer types:
```python
>>> import datatable as dt
>>> import numpy as np
>>> NP = np.array([0], dtype=np.uint8)
>>> DT = dt.Frame(NP)
ValueError: Unknown format 'B' with itemsize 1
```

There is an issue to add unsigned datatable column types https://github.com/h2oai/datatable/issues/1628 that could solve this problem.

We could also upcast `np.uint8`/`np.uint16`/`np.uint32` to datatable's `int16`/`int32`/`int64`. However, upcast won't help for `np.uint64`.