Make typed struct construction trim-friendly

[quinnj]

Summary

• route struct-shaped construction through a compact FieldSink{T}
• use a concrete Val{T} token and a bounded generated field assignment ladder
• preserve ordinary constructors, custom styles, hooks, defaults, tags, and mutable construction
• add an opt-in ordered-source cursor without exposing JSON-specific key behavior
• apply inbound ignore=true tags consistently during construction
• raise the package version to 2.8.3 for the companion JSON.jl integration

Why

This keeps typed construction statically reachable for trim compilation without adding a second interpreter, a package-specific hot registry, or per-struct annotations. It still uses normal Julia specialization per target type. Applications can persist those specializations with the standard PrecompileTools.@compile_workload pattern.

This is a smaller alternative to #63. The companion JSON.jl change is JuliaIO/JSON.jl#475. It requires StructUtils 2.8.3 and should land after this PR.

Performance

The coupled JSON/StructUtils audit used Julia 1.12.6 and one pinned Parsers revision across every variant.

• 100 mixed DTO first-use time: 9.78 seconds versus 16.48 seconds upstream
• standard workload precompile: 16.44 seconds and a 10.1 MB DTO cache
• the same old :hot workload: 58.07 seconds and a 40.5 MB DTO cache
• 100 DTO parses after restart: 2.1 milliseconds with the standard workload
• annotation-free 100-DTO trim executable: 9.96-second build, 3.9 MB binary, and successful execution

Small typed structs remain 10-16% slower than the current JSON/StructUtils pair. Bulk vectors of small structs remain 16-24% slower. Larger nested and ordered-wide cases improve by 3-31x. These tradeoffs are documented for review rather than hidden in an aggregate score.

Validation

• StructUtils construction, defaults, tags, custom styles, hooks, mutable targets, and table extensions
• inbound-ignore behavior and alias handling
• coupled JSON public API and semantic matrix
• safe-trim construction lanes with zero per-struct annotations
• 100-type compile-scaling and precompile-persistence tests

Full downstream application trim validation remains a pre-merge integration gate for the coupled change.

Co-authored by Codex