Eliminate all Core.Box captures to reduce latency and allocations

[hz-xiaxz]

Description

This PR addresses a significant number of Core.Box captures throughout the codebase. These were identified using the scanning script from JuliaLang/julia#60478.
Core.Box occurs when the compiler cannot infer the type or lifetime of a variable captured in a closure (often due to reassignment), forcing it onto the heap. This causes:
Type instabilities (inhibiting optimizations like SIMD).
Unnecessary allocations (increasing GC pressure).
Latency (compilation overhead).
While this PR touches many files, the changes are largely mechanical. I hope it won't be a large burden to review.

Strategies Used

To eliminate boxing, I employed several standard techniques depending on the context:

1. SSA-style Renaming: Ensuring variables captured by closures are not reassigned (e.g., x = ...; x = ... changed to x_init = ...; x_final = ...).
2. Ref containers: Using Ref for variables that genuinely need to be mutated inside closures (e.g., buffers or counters), keeping the binding constant.
3. Explicit Loops: Replacing map with explicit for loops where stateful iteration made closures inefficient.
4. Lifting Functions: Moving named inner functions to the module scope where they don't capture local variables.
5. Ordering: Fixing forward-reference issues by declaring variables before their capturing closures.

The scan results show a drastic reduction in boxed variables.

Before (Detected Boxes)

Var	Func	Location
pvmf32	#553	projection_utils.jl:276
viewport	#Scene#303	scenes.jl:236
labels	#get_labeled_plots	legend.jl:1014
lplots	#get_labeled_plots	legend.jl:1014
pos	#initialize_block!	textbox.jl:162
relayout	#initialize_block!	legend.jl:42
i	#initialize_block!	menu.jl:1
was_inside_button	#initialize_block!	menu.jl:1
was_inside_options	#initialize_block!	menu.jl:1
last_task	#on_latest	observables.jl:26
run_f	#on_latest	observables.jl:26
N	#plot!	contours.jl:180
mesh_idx	#plot!	poly.jl:183
xy	#position_on_plot	ray_casting.jl:378
nrows	#print_columns	recipes.jl:709
new_theme	#set_theme!	theming.jl:209
ticklabels	LineAxis	lineaxis.jl:264
fn	String	text.jl:874
result_type	_register_argument_conversions!	compute-plots.jl:491
x	_register_argument_conversions!	compute-plots.jl:491
align	add_ticks_and_ticklabels!	axis3d.jl:592
maxi	aggregation_implementation!	datashader.jl:143
mini	aggregation_implementation!	datashader.jl:143
i	attribute_per_char	text_layouting.jl:9
bar_labels	barplot_labels	barplot.jl:262
deregister	connect_conversions!	dim-converts.jl:121
tfontsize	draw_axis3d	axis.jl:226
cam	initialize_block!	axis3d.jl:3
default_format	initialize_block!	slidergrid.jl:8
extract_range_format	initialize_block!	slidergrid.jl:8
reset_to_stored	initialize_block!	textbox.jl:4
textplot	initialize_block!	textbox.jl:4
x1	line_rectangle_intersection	annotation.jl:920
x2	line_rectangle_intersection	annotation.jl:920
y1	line_rectangle_intersection	annotation.jl:920
y2	line_rectangle_intersection	annotation.jl:920
deregister	needs_tick_update_observable	dim-converts.jl:160
lastcomm	parse_bezier_commands	bezier.jl:401
buffer	plot!	timeseries.jl:39
nanpoint	plot!	band.jl:36
h	projectionmatrix	axis3d.jl:331
w	projectionmatrix	axis3d.jl:331
field	replay_events	event-recorder.jl:32
value	replay_events	event-recorder.jl:32
ind	streamplot_impl	streamplot.jl:134
c1	to_image	patterns.jl:126
c2	to_image	patterns.jl:126
bb	unchecked_boundingbox	text_boundingbox.jl:28
update_plotlist	update_plotspecs!	specapi.jl:666
color	voxel_colors	voxels.jl:282

After: No

We can add a Test.detect_closure_boxes in tests to avoid introducing more boxes.

Type of change

Delete options that do not apply:

• Performance?

AI Disclose

I used some AI to assist me, but checked manually.

Checklist

• Added an entry in CHANGELOG.md (for new features and breaking changes)
• Added or changed relevant sections in the documentation
• Added unit tests for new algorithms, conversion methods, etc.
• Added reference image tests for new plotting functions, recipes, visual options, etc.

[hz-xiaxz]

Motivated by the performance bump, I continued to eliminate all Core.Box found by JETLS.