[nicoburns]

LLM Contributions: Written by Devin (Cognition AI) under direction of @nicoburns.

Rebase of #46 / #32 onto main (3fdcdca, which includes upstream linebender#790 conditional hanging). Supersedes both. Recovers most of the calculate_content_widths regression from linebender#715 on a 250 KB real-text corpus:

build	median ms	min ms
v0.11.0 (flat ClusterData scan)	0.68–0.70	0.66
main (3fdcdca)	1.14–1.36	1.12
this PR	0.77–0.88	0.75

CPU-pinned, interleaved, n=12 x 3; widths-result identical (min: 1160.2813, max: 17458.617).

Commit 1 — engine: ShapedClusterFlags has 5 free bits; they now hold the first character's Boundary (2 bits) and Whitespace (3 bits — all 8 variants), set in process_shaped_clusters::flush from first_character.info (already loaded there). Additive API on ShapedCluster: boundary_before(), whitespace(), char_len(). ShapedCluster stays 20 bytes. (The commit message mentions the layout use; that lives in commit 2.)

Commit 2 — layout (+ engine round-trip test for the flag decode): when the run has no letter/word spacing (the common case), calculate_content_widths scans shaped_clusters directly instead of atoms_start():

• break/newline checks only on atom-start clusters (i == 0 || is_grapheme_start()); a skip_atom flag skips a Newline atom's continuation clusters exactly like the atom loop's continue;
• hanging accumulator updated per cluster (if cluster_hangs { += advance } else { = 0 }) — with no gaps this is equivalent to atom_hanging_advance's "hangs entirely / hangs partially from the logical end" rule;
• whitespace_hangs(first char, style) is checked before char_len() == 1, so letters never load chars_range; multi-char clusters check every character with its own style, as atom_hanging_advance does;
• hangs_conditionally is set from the cluster's style; the atom loop uses the atom's first-character style, which only differs for mixed-style multi-cluster atoms (the edge case already flagged in Implement CSS's conditional hanging of whitespace linebender/parley#790 review).

Runs with spacing keep the existing atoms_start() loop (it needs Atom for spacing.atom_advance / atom_hanging_advance).

A measurement-only variant with no hanging logic at all runs at ~0.60 ms, so the remaining gap to 0.11 is the hanging-whitespace semantics 0.11 didn't have, not the split data model.

Changelog: None

Link to Devin session: https://dioxus.staging.devinenterprise.com/sessions/2564f38c1f574d04ac6e9a57a7284d93
Open in Devin Desktop: https://dioxus.staging.devinenterprise.com/desktop/session/2564f38c1f574d04ac6e9a57a7284d93?variant=devin-insiders

[staging-devin-ai-integration]

I'll fix CI failures and address comments from users with write access that start with 'DevinAI' or '@devin'.

• Disable automatic comment, CI, and merge conflict monitoring