How to avoid upcast for columns with missing values when creating dataframe

I want to create a pandas dataframe from a json string. To be able to flatten the nested json source data I need to use pandas.json_normalize() (instead of pandas.read_json()). In the source data some of the columns are integer type and sporadically have missing values. This causes the column to be "upcast" to float:

import pandas as pd

data = [
  { "c1": 1, "c2": 1, "c3": 1.01 },
  { "c1": 2, "c3": 2.02 },
  { "c1": 3, "c2": 3, "c3": 3.03 }
]

df = pd.json_normalize(data)
print(df, "\n\n", df.dtypes)

which results in

   c1   c2    c3
0   1  1.0  1.01
1   2  NaN  2.02
2   3  3.0  3.03 

c1      int64
c2    float64
c3    float64

(I have stripped down the example to the crux of the matter. In reality the source data is far more complex.)

Applying df.convert_dtypes() kind of solves the problem after the fact but the function feels a bit creepy to me because I don't really know what it does to the other columns that I don't want to change (such as datetime columns).

Is there a way to avoid this "upcast" in the first place without affecting other columns?