Stop auto-converting timezones to UTC and stripping timezone info

[jonmmease]

Overview

This PR introduces a new v4 timezones future flag that changes the behavior of datetime serialization in plotly.py.

The current (version 3) behavior for datetimes with timezone info is for plotly.py to automatically convert datetimes to UTC and then remove the timezone info during serialization. This has resulted in a lot of confusion over the years (See #209).

The new behavior is for plotly.py to stop performing the UTC conversion and leave the timezone info in the string that is passed along to plotly.js. Plotly.js currently drops the timezone info (plotly/plotly.js#1532) and displays everything in local time, but hopefully at some point in the future Plotly.js will process the timezone info and provide a way to specify the desired display timezone in the figure specification.

Example

v3 behavior

from _plotly_future_ import renderer_defaults
import plotly.graph_objs as go
import plotly.io as pio
import pandas as pd
import sys, datetime

x1 = pd.to_datetime(
        ['2015-04-04 12:00:00', '2015-04-04 13:00:00', '2015-04-04 15:00:00'])
x2 = x1.tz_localize('UTC').tz_convert('US/Eastern')

fig1 = go.Figure(data=[go.Scatter(x=x2, y=[1, 3, 2])])
fig1

print(pio.to_json(fig1, pretty=True))

{
  "data": [
    {
      "type": "scatter",
      "x": [
        "2015-04-04 12:00:00",
        "2015-04-04 13:00:00",
        "2015-04-04 15:00:00"
      ],
      "y": [
        1,
        3,
        2
      ]
    }
  ],
  "layout": {}
}

v4 behavior

from _plotly_future_ import renderer_defaults, timezones
import plotly.graph_objs as go
import plotly.io as pio
import pandas as pd
import sys, datetime

x1 = pd.to_datetime(
        ['2015-04-04 12:00:00', '2015-04-04 13:00:00', '2015-04-04 15:00:00'])
x2 = x1.tz_localize('UTC').tz_convert('US/Eastern')

fig1 = go.Figure(data=[go.Scatter(x=x2, y=[1, 3, 2])])
fig1

print(pio.to_json(fig1, pretty=True))

{
  "data": [
    {
      "type": "scatter",
      "x": [
        "2015-04-04T08:00:00-04:00",
        "2015-04-04T09:00:00-04:00",
        "2015-04-04T11:00:00-04:00"
      ],
      "y": [
        1,
        3,
        2
      ]
    }
  ],
  "layout": {}
}

cc @nicolaskruchten @chriddyp @alexcjohnson

[nicolaskruchten]

This seems like the right thing to do overall: match the plotly.js API more closely such that the day plotly.js handles timezones more gracefully it'll just flow through.

[jonmmease]

To try out the dev build:

$ pip install https://13363-14579099-gh.circle-artifacts.com/0/dist/plotly-3.9.0%2B9.gdc776963.tar.gz

Then start your python notebook/script with

from _plotly_future_ import timezones