Why is read_csv so much slower to parse dates than to_datetime

Question

I am using jupyter notebook and reading a .csv file with pandas read_csv. When using the following code, it takes a reeeeeeeally long time (more than 10 minutes). The dataset has 70821 entries.

    [Input]
    df = pd.read_csv("file.csv", parse_dates=[0])
    df.head()
    
    [Output]
    created_at  field1  field2
    2022-09-16T22:53:19+07:00   100.0   NaN
    2022-09-16T22:54:46+07:00   100.0   NaN
    2022-09-16T22:56:14+07:00   100.0   NaN
    2022-09-16T23:02:01+07:00   100.0   NaN
    2022-09-16T23:03:28+07:00   100.0   NaN

If I just use parse_dates=True it will not detect the date and keep it as object.

If I read the dataset without parse_dates it goes much faster like I would expect it (~1 second). When I then use a seperate line of code to parse the date like

df["created_at"]=pd.to_datetime(df['created_at'])

it goes faster than using parse_dates in read_csv but still takes couple of minutes (around 3 minutes).

Using the following

df["created_at"]=pd.to_datetime(df['created_at'], format="%Y-%m-%d %H:%M:%S%z")

or with the T in the format string

df["created_at"]=pd.to_datetime(df['created_at'], format="%Y-%m-%dT%H:%M:%S%z")

or

df["created_at"]=pd.to_datetime(df['created_at'], infer_datetime_format=True)

does not increase the speed (still around 3 minutes)

So my questions are the following:

1. Why is the slowest way to parse the date directly with read_csv?
2. Why is the way with pd.to_datetime faster and
3. Why does something like format="%Y-%m-%d %H:%M:%S%z" or infer_datetime_format=True not speed up the process
4. And finally how do I do it in a better way? If I read the file once and parsed it to datetime, what would be the best way to write it back to a csv file so I don't have to go through this process all over again? I assume I have to write a function and manually change every entry to a better formatted date?

Can someone help me figuring out why those different approaches take such a different amount of time and how I speed up e.g. with something I tried in 3.? Thanks a lot.

---

EDIT:

I tried now to manually adjust the date format and see, where it causes trouble. Turns out, when I delete +07:00 in the date string, it is fast (~500 ms).

Under this following link I uploaded to csv files. example1 is the file with the problematic datetime format. In example2_no_timezone I deleted in every entry the +07:00 part which makes the parsing fast again (expected behaviour).

Folder with two example datasets

The questions above do remain sadly.

1. Why is pandas not able to read the original date string in an appropriate time
2. Why is to_datetime faster (but still too slow with the original dataset
3. How do I fix this without changing the format in the original dataset (e.g., with means of to_datetime and providing format=...)