BUG: convert_dtypes raises OverflowError for integers outside the int64 range #66517

Pandas version checks

• I have checked that this issue has not already been reported.

• I have confirmed this bug exists on the latest version of pandas.

• I have confirmed this bug exists on the main branch of pandas.

Reproducible Example

import pandas as pd

pd.Series([2**64], dtype=object).convert_dtypes()
# OverflowError: Python int too large to convert to C long

pd.Series([-(2**63) - 1], dtype=object).convert_dtypes()
# OverflowError: Python int too large to convert to C long

Issue Description

Series construction correctly infers object for an integer that no NumPy/nullable integer
dtype can hold:

>>> pd.Series([2**64]).dtype
dtype('O')

but convert_dtypes then tries to force that column to Int64 regardless, and overflows:

  File "pandas/core/internals/blocks.py", line 582, in convert_dtypes
    rbs.append(b.astype(dtype=dtype, squeeze=b.ndim != 1))
  File "pandas/core/dtypes/astype.py", line 82, in _astype_nansafe
    return dtype.construct_array_type()._from_sequence(arr, dtype=dtype, copy=copy)
  File "pandas/core/arrays/numeric.py", line 248, in _coerce_to_data_and_mask
    values = dtype_cls._safe_cast(values, dtype, copy=False)
  File "pandas/core/arrays/integer.py", line 65, in _safe_cast
    casted = values.astype(dtype, copy=copy)
OverflowError: Python int too large to convert to C long

lib.infer_dtype reports "integer" for these values, and the object-block conversion maps
that straight to Int64 without checking that the values actually fit. IntegerArray._safe_cast
first attempts astype(..., casting="safe"), which raises TypeError for object input, and the
fallback values.astype(dtype) is what raises OverflowError.

Note this is a plain OverflowError escaping a public API rather than a pandas error type, so
it is also not catchable via anything in pandas.errors.

Expected Behavior

convert_dtypes should leave the column as object when the values do not fit in the target
integer dtype, matching what construction already infers:

>>> pd.Series([2**64], dtype=object).convert_dtypes().dtype
dtype('O')

Raising a clear pandas-level error would also be defensible, but silently attempting Int64 and
surfacing a raw OverflowError is not.

Installed Versions

INSTALLED VERSIONS
------------------
commit                : e68db09ecf6427d1b62e565bacf17f2e525a3032
python                : 3.13.11
python-bits           : 64
OS                    : Darwin
OS-release            : 23.3.0
pandas                : 3.0.5
numpy                 : 2.5.1
dateutil              : 2.9.0.post0