feat(layout): expose positioned visual cluster iteration

[ldm0]

Moli currently has to reproduce Parley's visual item positioning to avoid calling the linear-time Cluster::visual_offset() for every cluster. Add Line::positioned_visual_clusters(), yielding (Cluster, f32) in visual left-to-right order with no position cache or heap allocation, so consumers can obtain all positions in one pass.

Offsets include LineMetrics::offset and exclude inline_min_coord, matching visual_offset() bit for bit. Preceding runs use run.advance() while clusters within a run use cluster.advance(), preserving spacing, justification and floating-point summation order. The iterator reads Parley's current visual item stream directly, including inline boxes and host item reordering, and supports cloning and iterator adaptors.

Validation:

• cargo fmt --all
• cargo clippy --workspace --all-targets --all-features -- -D warnings
• cargo nextest run --no-fail-fast: 502 passed, 1 skipped.
• RUSTDOCFLAGS='--cfg docsrs -D warnings' cargo +nightly doc --workspace --all-features --no-deps --document-private-items
• cargo check -p parley --no-default-features --features libm
• Four regression tests covering mixed bidi text, ligatures, combining characters, positive/negative letter spacing, word spacing, consecutive and zero-width inline boxes, indent, all six alignments, repeated line breaking/alignment, nonzero line origins, visual box reordering, clone/nth/skip/last, empty/box-only lines and an 8,192-cluster single run. Positions are checked against the independent visual_offset() implementation, using exact float bits.

Added release benchmarks comparing the two library access patterns on already-built layouts. Final revision a18cfc3ba6638d44ff0d0c9ce7b0b3396b42e8cf, Intel Core i9-13900K, rustc 1.94.0. Three separate process runs, 100 samples per case; the table reports the median of the three process means as rounded by Tango:

Text	Clusters	Per-cluster visual_offset()	Positioned iterator
LTR single run	128	146.6 µs	2.3 µs
LTR single run	512	2.3 ms	8.8 µs
LTR single run	2,048	36.1 ms	35.8 µs
Mixed LTR/RTL	128	20.0 µs	2.8 µs
Mixed LTR/RTL	512	174.7 µs	11.1 µs
Mixed LTR/RTL	2,048	2.2 ms	44.6 µs

Reproduce with cargo +1.94.0 bench -p parley_bench --bench main -- solo --filter 'Cluster Positions -*' --samples 100 --seed 20261016.

These measurements compare Parley's public API access patterns. Moli already has a linear accumulator after lexmount/moli#1434; the integration consolidates ownership of that existing algorithm.

The base is fix/inline-object-bidi-rebased at d0f5452487b872562102fde2c133d445272bcccc, the exact revision currently used by Moli, to avoid pulling unrelated fork changes into the integration.

Moli consumer integration: lexmount/moli#1599.