Use local time for column/row headers and computed functions

[sc1f]

This PR fixes several issues with computed functions and computed columns in perspective-python:

• the hour_of_day, day_of_week, and month_of_year functions now use local time to perform their calculations. Previously, all calculations were in UTC which could cause issues with datetimes that were on the cusp of one day/month to the next, as well as returning the wrong hour of day when compared with the input column.
• row and column headers for datetimes (i.e. when a datetime column is pivoted on) are now in local time, consistent with the behavior of datetimes across Perspective:

• Additional tests have been added for both features
• Edits to documentation around datetime handling in the Python guide
• Fixes a race condition where computed columns created via attribute were created but not preserved after the view changes.
• Fixes a scrolling issue within Jupyterlab (and other environments where Perspective is both > page width and the page is scrollable vertically) where using the arrow keys to navigate autocomplete would move the entire page.

[andyferris]

Is it possible to have both behaviours implemented and be able to choose?

[sc1f]

Is it possible to have both behaviours implemented and be able to choose?
@andyferris

Which behaviours are you referring to exactly? For the computed functions, even though it would be possible to preserve the old behaviour of UTC, it just isn't correct when compared with the underlying data:

Before

After

The issue is the same with row headers - we don't have an API for controlling their output from userland (they are generated in the C++ part of the codebase), and UTC output would be inconsistent with the underlying data.

[texodus]

Thanks for the PR! This is a fantastic fix.

@andyferris With this change you can present datetime values in whatever timezone and format you choose, either via perspective-viewer-datagrid formatting, or the localized datetime from the perspective API. I think the important take away is that time-zone relativeness in encoding is now preserved, which is very important as perspective data on the server and client may be distributed across timezone. If you have any questions about this or have suggestions for how to improve this API, I encourage you to open a new Issue to describe it - we are very interested in feedback on this topic!