Fix adding shapes and annotations inside batch_update() - #5758

Link to issue

Closes #4742

Description of change

Inside batch_update() property edits are deferred until the context exits, but appending to a compound array property (fig.layout.shapes += (shape,)) still swaps in the new child objects right away. Those objects have no properties to read back until the batch is applied. add_vline(), add_hline(), add_vrect() and add_hrect() read the shape they just added to append " domain" to its axis reference, which gives the TypeError: unsupported operand type(s) for +=: 'NoneType' and 'str' from the issue. The same thing makes a second add_shape()/add_annotation() inside one batch copy the first object as {}, so it silently disappears.

Adding layout objects is a structural change like adding traces, and add_traces() already applies those immediately regardless of batch mode. This does the same for layout objects: _add_annotation_like() appends the object, and _process_multiple_axis_spanning_shapes() adds and fixes up the axis-spanning shapes, with batch mode switched off (new _batch_mode_disabled() context manager next to batch_update()). Regular property edits in the batch are still deferred and sent in one update message as before.

Demo

import plotly.graph_objects as go

fig = go.Figure()
with fig.batch_update():
    fig.add_vline(1)
fig.layout.to_plotly_json()["shapes"]

before:

TypeError: unsupported operand type(s) for +=: 'NoneType' and 'str'

after (same as without batch_update()):

[{'type': 'line', 'x0': 1, 'x1': 1, 'xref': 'x', 'y0': 0, 'y1': 1, 'yref': 'y domain'}]

Two shapes in one batch:

fig = go.Figure(go.Scatter(x=[1, 2], y=[1, 2]))
with fig.batch_update():
    fig.add_shape(type="line", x0=0, x1=1, y0=0, y1=1)
    fig.add_shape(type="rect", x0=1, x1=2, y0=1, y1=2)
fig.layout.to_plotly_json()["shapes"]

before:

[{}, {'type': 'rect', 'x0': 1, 'x1': 2, 'y0': 1, 'y1': 2}]

after:

[{'type': 'line', 'x0': 0, 'x1': 1, 'y0': 0, 'y1': 1}, {'type': 'rect', 'x0': 1, 'x1': 2, 'y0': 1, 'y1': 2}]

Testing strategy

test_add_axis_spanning_shape_in_batch_update (autoshapes tests) adds each of the four shapes, some with annotations, to a single plot, to one subplot and to all subplots, inside and outside batch_update(), and checks that the resulting shapes and annotations are identical. test_add_layout_objects_in_batch_update (figure messages tests) checks that shapes and annotations added inside a batch are there right away and go out as their own relayout messages, while a regular property assignment in the same batch is still deferred and sent in the single update message. All 13 new cases fail on main. Running tests/test_core, test_autoshapes, test_px and test_figure_factory with and without the change gives the same set of other failures (Windows cp1252 decoding in test_offline, statsmodels and scikit-image not installed here).

Additional information (optional)

I kept the fix local to adding layout objects. Assigning a compound array property directly inside a batch (fig.layout.shapes = [...]) still leaves the new children without readable properties until the batch is applied; changing that would touch how FigureWidget batches its messages, so I left it alone here.

Guidelines

• I have reviewed the pull request guidelines and the Code of Conduct and confirm that this PR follows them.
• I have added an entry to the changelog if needed (not required for documentation PRs).