[andrewchen5678]

The two planned items of the roadmap's Lossless tiles, and release 0.0.120.

• The moving frames' zstd, faster. A frame of its own (KIND_PLANAR) now steps 8 bytes past a place the fast strategy found no match at (targetLength under ZSTD_fast), its literals forced to stay Huffman coded (literalCompressionMode, behind the zstd crate's experimental feature, which the bundled libzstd source already carries). The still stream keeps every look.
• The still stream's table. 2^16 entries and an 8 MiB window, from 2^12 and 1 MiB: the small table forgot a match further back than a few thousand places, so a larger window had taken nothing off. 8 MiB is the most the protocol lets a STILL's frame have, and zDesk's page already holds it to exactly that (STILL_WINDOW_MOST), so neither the protocol nor the page changes. The test reader now holds the window to 8 MiB as the page does.

Measured on lab1, alone, motion-bench --tiles-only --lossless over every sample at every size, two runs each, against 0.0.119:

	0.0.119 → now
CPU, all but flood	197 → 170 s (0.87×)
video at 1920x1080	11.6 → 9.9 ms a frame
video at 4K	45.9 → 39.2 ms a frame
bytes, all but flood	15,064 → 15,027 MB
browser	19.6 → 3.6 MB
terminal / uncover / still	0.68× / 0.69× / 0.45×

The still table costs the mixed session 4% more of the processor; moving pictures carry up to 0.4% more bytes. Figures in docs/tiles-performance.md, design in docs/architecture.md; the bench is archived as bench/lossless-zstd-20261105/. wlshare-proto is unchanged, so no proto-v tag.

cargo test, cargo clippy --all-targets -- -D warnings and cargo fmt --check pass.

🤖 Generated with Claude Code