BUG/TST: assure conversions of datetimelikes for object, numeric dtypes

[jreback]

closes #19223
closes #12425

[jreback]

This is fixed as well (and present in <= 0.22.0), I thought we had an issue about this

In [1]: pd.DataFrame({'A': [1,2,3]}).astype('m8[s]').dtypes
Out[1]: 
A    timedelta64[s]
dtype: object

[codecov]

Codecov Report

Merging #19224 into master will decrease coverage by <.01%.
The diff coverage is 88.88%.

@@            Coverage Diff             @@
##           master   #19224      +/-   ##
==========================================
- Coverage   91.53%   91.52%   -0.01%     
==========================================
  Files         147      147              
  Lines       48797    48797              
==========================================
- Hits        44664    44661       -3     
- Misses       4133     4136       +3

Flag	Coverage Δ
#multiple	89.89% <88.88%> (-0.01%)	⬇️
#single	41.6% <55.55%> (ø)	⬆️

Impacted Files	Coverage Δ
pandas/core/internals.py	94.5% <100%> (ø)	⬆️
pandas/core/dtypes/common.py	94.58% <100%> (+0.04%)	⬆️
pandas/core/dtypes/cast.py	88.35% <75%> (-0.31%)	⬇️
pandas/util/testing.py	84.2% <0%> (-0.22%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 08a3d3a...b953695. Read the comment docs.