BUG: calling Timestamp constructor on Timestamp at DST changes Timestamp value

[AlexKirko]

Code Sample, a copy-pastable example if possible

tz = "dateutil/Europe/London"
ts = pd.Timestamp("2013-10-27 01:00:00")
print("Initial value is {}".format(ts.value))
ts = ts.tz_localize(tz, ambiguous=False, nonexistent="raise")
print("After localization value is {}".format(ts.value))
ts = pd.Timestamp(ts)
print("After casting Timestamp as Timestamp value is {}".format(ts.value))

Output:
0.26.0.dev0+1608.g06c5d2488

Initial value is 1382835600000000000
After localization value is 1382835600000000000
After casting Timestamp as Timestamp value is 1382832000000000000

Problem description

TLDR: Calling the Timestamp constructor on a Timestamp shouldn't change the object in any way.

Long version: Ran into this while working on solving #30543. When we localize into a Daylight Savings Time timezone, we are forced to alter the underlying value of a Timestamp to make it fit non-DST timezones. WIthout this, date arithmetic won't work properly between timezones. This is also necessary, for example, to make sure that a pd.date_range with one of the ends on a DST shift doesn't break (the test for this is implemented by test_dti_construction_ambiguous_endpoint in pandas.tests.indexes.datetimes.test_timezones.TestDatetimeIndexTimezones). Unfortunately, what happens currently is that ts.tz_localize doesn't change the underlying value on a DST shift. It instead changes when the Timestamp constructor is called on a localized Timestamp (you can take a look, for example, here).

Expected Output

Initial value is 1382835600000000000
After localization value is 1382832000000000000
After casting Timestamp as Timestamp value is 1382832000000000000

The value should change on localization, not on calling the constructor.
I'd like to work on a PR, because without fixing this it's impossible to solve #30543 without introducing ugly hacks.

xref: #30543

[mroeschke]

Thanks for the report. I think this is a duplicate of #24329