Regression in to_timedelta with errors="coerce" and unit

[TomAugspurger]

Introduced in d3f686b

In pandas 1.0.3

In [2]: pd.to_timedelta([1, 2, 'error'], errors="coerce", unit="ns")
Out[2]: TimedeltaIndex(['00:00:00.000000', '00:00:00.000000', NaT], dtype='timedelta64[ns]', freq=None)

In master, we raise.

In [2]: pd.to_timedelta([1, 2, 'error'], errors="coerce", unit="ns")
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-2-a3691c044041> in <module>
----> 1 pd.to_timedelta([1, 2, 'error'], errors="coerce", unit="ns")

~/Envs/dask-dev/lib/python3.7/site-packages/pandas/core/tools/timedeltas.py in to_timedelta(arg, unit, errors)
    101         arg = arg.item()
    102     elif is_list_like(arg) and getattr(arg, "ndim", 1) == 1:
--> 103         return _convert_listlike(arg, unit=unit, errors=errors)
    104     elif getattr(arg, "ndim", 1) > 1:
    105         raise TypeError(

~/Envs/dask-dev/lib/python3.7/site-packages/pandas/core/tools/timedeltas.py in _convert_listlike(arg, unit, errors, name)
    140
    141     try:
--> 142         value = sequence_to_td64ns(arg, unit=unit, errors=errors, copy=False)[0]
    143     except ValueError:
    144         if errors == "ignore":

~/Envs/dask-dev/lib/python3.7/site-packages/pandas/core/arrays/timedeltas.py in sequence_to_td64ns(data, copy, unit, errors)
    927     if is_object_dtype(data.dtype) or is_string_dtype(data.dtype):
    928         # no need to make a copy, need to convert if string-dtyped
--> 929         data = objects_to_td64ns(data, unit=unit, errors=errors)
    930         copy = False
    931

~/Envs/dask-dev/lib/python3.7/site-packages/pandas/core/arrays/timedeltas.py in objects_to_td64ns(data, unit, errors)
   1037     values = np.array(data, dtype=np.object_, copy=False)
   1038
-> 1039     result = array_to_timedelta64(values, unit=unit, errors=errors)
   1040     return result.view("timedelta64[ns]")
   1041

pandas/_libs/tslibs/timedeltas.pyx in pandas._libs.tslibs.timedeltas.array_to_timedelta64()

ValueError: unit must not be specified if the input contains a str

This restores the 1.0.3 behavior, and adds an additional test for errors="ignore", and cleans up the to_timedelta docstring.

[jreback]

thanks @TomAugspurger