`px.timeline` with a Daylight Savings Time scenario - missing/extra hour

Problem:
When using px.timeline on a range where a Daylight Savings Time switch occurs & time zones are in the timestamps, I observe an unexpectedly shorter/longer duration of the bar.

Example A - March - missing an hour

import pandas as pd
import plotly.express as px

df = pd.DataFrame(dict(
    x_start=[
        pd.Timestamp("2024-03-09", tz="US/Pacific"),
        pd.Timestamp("2024-03-10", tz="US/Pacific"),
        pd.Timestamp("2024-03-11", tz="US/Pacific"),
    ],
    x_end=[
        pd.Timestamp("2024-03-10", tz="US/Pacific"),
        pd.Timestamp("2024-03-11", tz="US/Pacific"),
        pd.Timestamp("2024-03-12", tz="US/Pacific"),
    ],
    y=[1, 1, 1],
))

px.timeline(df, x_start="x_start", x_end="x_end", y="y")

As you can see, for the DST day (March 10, 2024), the end date is unexpectedly 1 hour early, unlike the other days.

Example B - November - extra hour

import pandas as pd
import plotly.express as px

df = pd.DataFrame(dict(
    x_start=[
        pd.Timestamp("2023-11-05", tz="US/Pacific"),
    ],
    x_end=[
        pd.Timestamp("2023-11-06", tz="US/Pacific"),
    ],
    y=[1],
))

px.timeline(df, x_start="x_start", x_end="x_end", y="y")

In this example, 1 hour is unexpectedly added to the end.

Testing environment:

• Google Colab (also experienced on VS Code & html)
• Latest version of plotly (5.22.0).

Commencez par reproduire les exemples pandas et px.timeline de l’issue, en couvrant les transitions DST de mars et de novembre. Suivez le point d’entrée de px.timeline dans sa gestion des horodatages avec fuseau horaire et ajoutez une couverture de régression pour les deux cas. Le travail est terminé lorsque chaque barre couvre l’intervalle calendaire local prévu sans heure manquante ni heure supplémentaire.