Cache primary font queries and font metrics across layouts in FontContext

[nicoburns]

LLM Contributions: This PR was written with Devin, which also ran the benchmarks below.

Stacks on #18 (devin/1788383056-vertical-align). #18 re-runs a fontique::Query per style per layout (resolve_style_metrics) and re-parses fonts for FontMetrics; the shaper's FontSelector repeats the same query per item. This adds a cross-layout cache owned by FontContext:

• Primary font cache: (FamilyId stack, FontWidth, FontWeight, FontStyle) -> Option<FontInstance + CharmapIndex>. Keyed on the actual FamilyId list because ResolveContext family-stack handles are per-LayoutContext interned indices, not stable across layouts. A miss runs the query but only accepts a font from the requested families (never from the query's fallback families), so the result is independent of fallback settings and the cached None means "no family font".
• Metrics cache: (blob id, font index, size bits, synthesis variation settings + style variations) -> FontMetrics.
• Both are bounded (MAX_ENTRIES = 1024, cleared when full).
• Invalidation: new fontique::Collection::generation() counter, bumped on every mutating Collection method (register/unregister fonts, load system fonts, load from paths, generic families, fallbacks, clear) and when a shared collection syncs a newer version from another clone. FontContext compares it before each layout and drops the cache on change. FontContext::clear_cache() exists for mutations the collection can't see (e.g. swapping blob data); FontContext::cache_len() exposes entry counts. Struct-literal construction of FontContext is replaced with FontContext::from_parts.
• The per-layout FontKey dedup in style_metrics.rs is kept as the first-level (hash-free) cache.
• Shaper: FontSelector takes the cached primary font for the current style. A cluster fully covered by it is resolved without touching the query (this is exactly what the query would yield first). Otherwise the normal per-cluster fallback query runs, seeded with the primary font's coverage so it isn't computed twice. Fallback behaviour is unchanged; emoji clusters bypass the cache entirely as before.
• Also adds the core_maths::CoreFloat import style_metrics.rs needed for --no-default-features --features libm (x86_64-unknown-none), which was failing on Implement CSS vertical-align for inline formatting contexts #18.

Tests: parley/src/tests/test_font_cache.rs covers reuse across layouts, invalidation on register_fonts / set_fallbacks, negative caching for unknown families, and Collection::generation() bumps.

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

Tango (parley_bench), base = #18 head ff83800, this branch = 5c31f79

cd parley-base && cargo export ../parley/target/benchmarks -- bench --bench=main   # at ff83800
cd parley/parley_bench && cargo bench -q --bench=main -- compare ../target/benchmarks/main -t 2

Benchmark	base ... this PR	Δ
Default Style - arabic 20 characters	[12.6 us ... 12.2 us]	-2.81%*
Default Style - latin 20 characters	[6.2 us ... 5.7 us]	-8.38%*
Default Style - japanese 20 characters	[11.4 us ... 11.5 us]	+0.98%
Default Style - arabic 1 paragraph	[67.3 us ... 65.6 us]	-2.49%*
Default Style - latin 1 paragraph	[24.2 us ... 21.5 us]	-11.39%*
Default Style - japanese 1 paragraph	[92.9 us ... 93.2 us]	+0.35%
Default Style - arabic 4 paragraph	[284.5 us ... 284.2 us]	-0.13%
Default Style - latin 4 paragraph	[91.8 us ... 82.1 us]	-10.62%*
Default Style - japanese 4 paragraph	[134.3 us ... 134.0 us]	-0.20%
Styled - arabic 20 characters	[15.0 us ... 14.5 us]	-3.26%*
Styled - latin 20 characters	[8.3 us ... 7.7 us]	-6.77%*
Styled - japanese 20 characters	[13.1 us ... 12.9 us]	-1.56%*
Styled - arabic 1 paragraph	[71.5 us ... 71.5 us]	-0.02%
Styled - latin 1 paragraph	[30.7 us ... 28.3 us]	-7.97%*
Styled - japanese 1 paragraph	[103.3 us ... 107.1 us]	+3.65%*
Styled - arabic 4 paragraph	[323.3 us ... 324.7 us]	+0.42%
Styled - latin 4 paragraph	[119.4 us ... 109.9 us]	-7.92%*
Styled - japanese 4 paragraph	[146.0 us ... 151.4 us]	+3.67%*
Word + Letter Spacing - arabic 20 characters	[12.4 us ... 12.3 us]	-0.27%
Word + Letter Spacing - latin 20 characters	[6.3 us ... 5.7 us]	-10.13%*
Word + Letter Spacing - japanese 20 characters	[12.1 us ... 11.7 us]	-2.88%
Word + Letter Spacing - arabic 1 paragraph	[67.1 us ... 66.0 us]	-1.62%*
Word + Letter Spacing - latin 1 paragraph	[24.4 us ... 22.0 us]	-9.98%*
Word + Letter Spacing - japanese 1 paragraph	[95.3 us ... 94.0 us]	-1.37%*
Word + Letter Spacing - arabic 4 paragraph	[287.7 us ... 288.0 us]	+0.12%
Word + Letter Spacing - latin 4 paragraph	[94.9 us ... 82.3 us]	-13.34%*
Word + Letter Spacing - japanese 4 paragraph	[133.8 us ... 133.6 us]	-0.11%
Repeated Justification - latin 4 paragraph	[1.0 us ... 984.3 ns]	-1.67%*
Caused by:
process didn't exit successfully: `/home/ubuntu/	epos/parley/target/release	deps/main-a3b52844cca21a67 compare ../target/benchmarks/main -t 2 --bench` (exit status: 1)

Latin (where the primary font covers everything) gains 5–13%, which more than recovers the +3–10% #18 cost on short styled layouts. Arabic/Japanese rows (primary Roboto covers nothing, so every cluster goes through fallback) are within noise except a ~+3.5% on two "Styled - japanese" rows that was not reproducible across three runs (−1.7%..+3.9%).

Changelog

Added

• fontique::Collection::generation() returns a counter that changes whenever the collection is mutated.
• parley::FontContext caches primary font selection and font metrics across layouts; FontContext::from_parts, FontContext::clear_cache, FontContext::cache_len.

Link to Devin session: https://dioxus.staging.devinenterprise.com/sessions/3a0b8cc210c041d8955c841fb2a76a8e
Open in Devin Desktop: https://dioxus.staging.devinenterprise.com/desktop/session/3a0b8cc210c041d8955c841fb2a76a8e?variant=devin-insiders
Requested by: @nicoburns

[staging-devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR that start with 'DevinAI' or '@devin'.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring