Reorder tests in maybe_downcast_numeric (#55825)

* Reorder tests in maybe_downcast_numeric

The comment `# if we have any nulls, then we are done` is not consistent with the test `if isna(arr).any()` because `arr` is constructed only from the first element (`r[0]`) not the full ravel'd list of values.  Moreover, calling `np.array()` on some random type can have surprising consequences.

So instead, do the early-out test as intended, just using `r[0]` without going through `np.array()`.  Then test other things about `r[0]`.  Only then should we test all the values (and if we have any nulls, then we are done).

See #55824

Signed-off-by: Michael Tiemann <[email protected]>

* Simplify/optimize tests for downcasting

If the first element of `result` is an array, ravel that to get element we will test.  Otherwise use it as is.

We only need to check whether `result` is all non-null once.

Signed-off-by: Michael Tiemann <[email protected]>

* Update cast.py

Don't use deprecated array indexing on ExtensionArrays.  We need to now us `iloc`.

Signed-off-by: Michael Tiemann <[email protected]>

* Eliminate need to call `ravel`

When processing a multidimensional `ndarray`, we can get the first element by calling `result.item(0)` and completely avoid the copying needed by `ravel` to get the first element that way.  We can also eliminates an additional conditional check.

Signed-off-by: Michael Tiemann <[email protected]>

---------

Signed-off-by: Michael Tiemann <[email protected]>
Co-authored-by: Richard Shadrach <[email protected]>

pandas/core/dtypes/cast.py

@@ -361,15 +361,11 @@ def trans(x):
             # if we don't have any elements, just astype it
             return trans(result).astype(dtype)
 
-        # do a test on the first element, if it fails then we are done
-        r = result.ravel()
-        arr = np.array([r[0]])
-
-        if isna(arr).any():
-            # if we have any nulls, then we are done
-            return result
-
-        elif not isinstance(r[0], (np.integer, np.floating, int, float, bool)):
+        if isinstance(result, np.ndarray):
+            element = result.item(0)
+        else:
+            element = result.iloc[0]
+        if not isinstance(element, (np.integer, np.floating, int, float, bool)):
             # a comparable, e.g. a Decimal may slip in here
             return result