Keep at least three screens of kitty images in the registry

[tomlm]

Problem

MaxImageRegistryBytes defaults to 32 MB and the registry trims the oldest images by id regardless of screen size. A client that tiles a full-screen picture (Consolonia's kitty renderer keeps up to three screens' worth of tiles in the terminal to show them again without resending) exceeds that from about 2560x1440 up, and a 4K screen alone is 33 MB. Once the registry has dropped a tile the client still counts on, placing it again silently shows nothing.

Change

The option becomes a floor: the kitty registry budget is max(MaxImageRegistryBytes, 3 × Cols × CellWidthPixels × Rows × CellHeightPixels × 4). Zero or less still means no limit, and a configured budget larger than three screens still applies. Only the two kitty store/charge sites use the registry budget for storage; the iTerm2 OSC sites keep using the option as a per-image sanity ceiling, so their tests at 1 and 1024 bytes are unchanged.

Three tests cover the floor, an image beyond it, and a larger configured budget.

🤖 Generated with Claude Code

[github-actions]

Perf comparison — this change, against its base

3 run(s) of each side, alternating on one machine. Allocation is a count and is gated exactly. Time is a measurement, so its gate is derived from the spread this job just observed in itself rather than fixed in advance.

corpus	bytes/char	gen0/Mchar	ns/char	Δ time	noise	gate
scroll-ascii	0.00 → 0.00	0.00 → 0.00	3.82 → 3.82	+0.0%	±8%	23%
sgr-churn	0.00 → 0.00	0.00 → 0.00	8.94 → 8.82	-1.4%	±12%	35%
truecolor	0.00 → 0.00	0.00 → 0.00	8.40 → 8.43	+0.3%	±2%	6%
alt-redraw	0.00 → 0.00	0.00 → 0.00	11.41 → 11.41	-0.0%	±1%	4%
unicode	7.66 → 7.66	0.09 → 0.09	30.45 → 30.28	-0.5%	±9%	26%
flood	0.00 → 0.00	0.00 → 0.00	117.76 → 117.33	-0.4%	±12%	35%

Each corpus is gated at max(4%, 3 × its own noise). A wide noise column means this runner was busy and the timing half of the table should be read as advisory; the allocation half is exact either way.

assemblies measured

• base: XTerm.NET 2.0.7.0 mvid:e66938d7-863d-4d2d-9d1c-d905837c5013
• head: XTerm.NET 2.0.7.0 mvid:86847d8b-d6b4-4891-a1ff-7390d4b7a377

---

Perf comparison — cumulative, everything since 2.0.3

3 run(s) of each side, alternating on one machine. Allocation is a count and is gated exactly. Time is a measurement, so its gate is derived from the spread this job just observed in itself rather than fixed in advance.

corpus	bytes/char	gen0/Mchar	ns/char	Δ time	noise	gate
scroll-ascii	0.00 → 0.00	0.00 → 0.00	3.89 → 3.82	-1.8%	±7%	21%
sgr-churn	0.00 → 0.00	0.00 → 0.00	8.98 → 8.82	-1.8%	±10%	30%
truecolor	0.00 → 0.00	0.00 → 0.00	8.42 → 8.43	+0.1%	±4%	13%
alt-redraw	0.00 → 0.00	0.00 → 0.00	11.45 → 11.41	-0.4%	±2%	5%
unicode	7.66 → 7.66	0.09 → 0.09	29.89 → 30.28	+1.3%	±10%	29%
flood	0.00 → 0.00	0.00 → 0.00	117.90 → 117.33	-0.5%	±14%	42%

Each corpus is gated at max(5%, 3 × its own noise). A wide noise column means this runner was busy and the timing half of the table should be read as advisory; the allocation half is exact either way.

assemblies measured

• base: XTerm.NET 2.0.3.0 mvid:9ba0ce98-0d1e-4424-8aa1-9ab012cdc07a
• head: XTerm.NET 2.0.7.0 mvid:86847d8b-d6b4-4891-a1ff-7390d4b7a377

[tomlm]

@copilot resolve the merge conflicts in this pull request