This PR removes host_view and open_view from the public API.

With the improvements in the speed to create a flat view in #1235, the difference between using a View created on a server and creating a View on the client using a handle to the server's table is minimal, if non-existent in most use-cases. This allows us to remove the host_view API, which doesn't fit in very well with our established public API semantics of "table as data container, view as query and data serializer."

'host_view' shoe-horns in the ability to access a View over the network so that we can use it to create a table, and makes the explanations around distributed/server/client mode more complicated. Because the old context implementation added a small overhead (scaling linearly with the size of the underlying data), host_view made sense for performance reasons.

With the removal of host_view and open_view, Perspective's remote API only allows the hosting and opening of Table objects. This makes it far easier to reason about, especially when it comes to the distinctions between distributed, server, and client mode while maintaining the same feature set in the public API.

Both distributed and server mode is easier to implement:

manager = PerspectiveManager()
manager.host_table("data_source", TABLE)

const worker = perspective.worker();
const websocket = perspective.websocket(URL);

// Open the table on the server - this is all that is needed for server mode
const remote_table = websocket.open_table("data_source");

// For distributed mode, create a view
const view = remote_table.view();

// Pass it into the JS table constructor
const client_table = worker.table(view);

It is also trivial to host subsets of a table based on already-created views:

# Table with large dataset that might be too big to transmit in one go
table = perspective.Table(big_dataset)

# Create a view on a subset of the data
filtered_view = table.view(filter=[["Sales", ">", 100]])

# Create a table using the filtered view
filtered_table = perspective.Table(filtered_view.to_arrow())

# Add an `on_update` callback to keep the tables in sync
def updater(port_id, delta):
    filtered_table.update(delta)

filtered_view.on_update(updater, mode="row")

# Easy to host the tables
manager.host_table("data_source", table)
manager.host_table("data_source_filtered", filtered_table)

Changelog

• Remove host_view and open_view from JS and Python
• Refactor Jupyterlab widget to create a new view instead of using open_view
• Fix regression in PerspectiveWidget.delete()to delete all views created on widget's manager instance before deleting table
• Update examples to use open_table and view creation
• Update documentation to include distributed/server mode in Python docs