Preserve unit from Pandas timestamp when creating arrow scalar

### Describe the enhancement requested

```
>>> ts = pd.Timestamp("2020-07-28 21:10:53.114429867")
>>> ts.unit
'ns'
>>> pa_s = pa.scalar(ts)
>>> pa_s.type
TimestampType(timestamp[us])
>>>
```

Since the pandas unit was nanoseconds, would be nice if that were preserved in the inferred type, without having to specify ns explicitly, as microseconds results in precision loss.

### Component(s)

Python

Reproduce the issue with the shown pa.scalar(ts) example and inspect the Python scalar conversion path used for pandas.Timestamp values. Confirm the inferred timestamp type preserves the pandas nanosecond unit and avoids precision loss, then add or update coverage for this example.

Written by the indexing model from the issue text.