Fix expand_dims string coordinate dtype inference

[garg-khushi]

This PR fixes an inconsistency where expand_dims created object-dtype
coordinates for string inputs instead of NumPy unicode dtype.

Changes:

• Preserve NumPy unicode dtype for string coordinates created via expand_dims
• Prevent object-dtype propagation through concat by improving
  PandasIndex coord dtype inference
• Add regression tests covering both cases

Fixes #11061

[jsignell]

Thanks for taking the time to work on this @garg-khushi and sorry it's taken a bit to review. I was wondering if there might be a simpler solution as mentioned in the comments. I tried out the change that I suggested locally and just had to tweak one test:

diff --git a/xarray/tests/test_dataset.py b/xarray/tests/test_dataset.py
index d25ef5a2..6f9edae4 100644
--- a/xarray/tests/test_dataset.py
+++ b/xarray/tests/test_dataset.py
@@ -3860,7 +3860,7 @@ class TestDataset:
         # Regression test for https://github.com/pydata/xarray/issues/7493#issuecomment-1953091000
         # todo: test still needed?
         ds = Dataset().expand_dims({"time": [np.datetime64("2018-01-01", "m")]})
-        assert ds.time.dtype == np.dtype("datetime64[s]")
+        assert ds.time.dtype == np.dtype("datetime64[m]")
 
     def test_set_index(self) -> None:
         expected = create_test_multiindex()

[jsignell]

I don't think we need any changes in this file actually.

[jsignell]

I'm not sure if we need to be quite so precise in only targeting objects. This seems to work just as well:

Suggested change

	coord_dtype = get_valid_numpy_dtype(index)
	if coord_dtype == object and index.dtype == object:
	inferred = getattr(index, "inferred_type", None)
	if inferred in ("string", "unicode"):
	coord_dtype = np.dtype(str)
	else:
	data = index.to_numpy(dtype=object, copy=False)
	if data.size and all(
	isinstance(x, (str, np.str_)) for x in data.ravel()
	):
	coord_dtype = np.asarray(data, dtype=str).dtype
	coord_dtype = get_valid_numpy_dtype(np.asarray(array))