Fullbleed 2.5.19: native Windows stack fix

Fullbleed 2.5.19 fixes native Windows stack overflows in unoptimized Rust document rendering. Container state and computed styles now keep their wide values on the heap, preserving the public container methods and rendering calculations.

The public crate passes the original small-table crash and nine standalone layout cases on Windows' unchanged 1 MiB main stack, including 16 nested blocks and three nested tables. No worker or linker stack override is used. The exact Rust quickstart example also compiles and renders against the public crate.

Validation includes 1,352 Rust library tests, 11 documentation tests, all 590 Python tests, and Windows/Linux CI. All 33 new regression PDFs and 48 counter PDFs match the preceding release. The complete unchanged source corpus retains 1,617 PASS, 26 FAIL, and 19 REFERENCE-DISPUTED, with all 1,662 PDF bytes unchanged. That corpus uses the explicitly identified pre-release source wheel; its existing failed gate remains in the evidence.

The added allocations have a modest cost in the local timing sample: about 1% for the designed invoice/report and 0.16 ms for the tiny table. These samples and finite nesting checks do not establish general performance, arbitrary-depth safety, or full CSS/PDF standards conformance. Node and .NET engine pins are unchanged.

• PyPI 2.5.19
• crates.io 2.5.19
• Rust API reference
• Native stack evidence and runnable registry consumer
• Complete source-corpus evidence
• Verification report
• SHA-256 checksums

Source: #63. Release preparation: #64. Resolves #62 after the public package checks recorded above.