BUG: Fix _replace_regex mutating StringArray in-place for non-inplace replace

[jbrockmendel]

This unblocks #57733

Block._replace_regex calls self.astype(np.dtype(object)) when the replacement value can't be held in the current dtype. For StringArray (backed by an object ndarray), this astype chain returns a block whose values share memory with the original StringArray._ndarray — because the underlying conversion is object→object with copy=False, which is a no-op. replace_regex then mutates this shared array in-place, corrupting the original.

Fix: call _maybe_copy(inplace=True) on the astype result before mutation, which copies only when the block shares refs with the original (i.e., when astype returned a view).

Reproducer:

df = pd.DataFrame({"b": list("ab..")})
df_orig = df.copy()
result = df.replace([r"\s*\.\s*", "b"], 0, regex=True)
# Before fix: df["b"]._ndarray contains ints, df != df_orig
# After fix: df is unchanged

🤖 Generated with Claude Code