Appending labelled points to a scatter puts them into the wrong class #1971

Description

Appending labelled points to a scatter plot puts them into the wrong class.

The client sends one trace per label present in the batch, named after the
label. The server ignores those names for an unnamed update and pairs the
entries with the plot's traces by position. So a batch holding only "dog"
is added to the first trace, "cat", and a label the plot hasn't seen yet is
added to an existing trace instead of getting its own.

Current Behavior

Starting from cat at x=1 and dog at x=2, 3:

append one dog (x=9)              -> cat=[1, 9]  dog=[2, 3]      wrong class
append a new class bird (x=7)     -> cat=[1, 7]  dog=[2, 3]      no bird trace
append bird (x=7) and dog (x=9)   -> cat=[1, 7]  dog=[2, 3, 9]   bird lost into cat

The same happens with integer labels that don't start at 1 (labels 3 and 7,
appending a 7 lands in "3").

Expected Behavior

Each appended point joins the trace of its own label, and a new label gets a
new trace.

Steps to Reproduce

import numpy as np
import visdom

viz = visdom.Visdom()
viz.scatter(X=np.array([[1., 1], [2, 2], [3, 3]]),
            Y=np.array(["cat", "dog", "dog"]), win="s")
viz.scatter(X=np.array([[9., 9]]), Y=np.array(["dog"]), win="s",
            update="append")
# the point at x=9 is in the "cat" trace

Environment

• visdom dev (reproduced at bda66a6)
• Python 3.12

The pairing is the zip(idxs, new_data) loop in UpdateHandler.update
(py/visdom/server/handlers/web_handlers.py).

FIX #1972