perf: batch lazy builtin initialization

[Eric-Song-Nop]

Stacked on #14. Base: perf/release-cleanup-fast-path; head: perf/batch-builtin-initialization, both in pocket-stack/quickjs-oxide. Review and merge in dependency order.

Migrated from Eric-Song-Nop#5 with identical commits and preserved measurement/validation evidence.

Context creation installed 121 Date/Array/TypedArray lazy methods individually, repeatedly replacing growing layouts. Install those methods in eight batches through the existing layout/ownership path, preserving property order, flags, metadata, lazy identity, realm ownership and alias boundaries.

将 121 项内建懒方法改为 8 批发布，复用现有布局与所有权机制，保留属性顺序、标志、函数元数据、realm 和别名关系。

• Context creation: 2.286008 → 1.756603 ms (−23.16%); layout replacements 778 → 665 per Context. / 初始化耗时与中间布局工作减少。
• Performance limitation: fixed property reads are 4.17% slower in the main series and 2.37% slower in six CPU-2-pinned follow-up pairs. Root cause remains unresolved; this is not a general interpreter speedup. / 固定属性负载存在已复测确认的回退，未宣称整体解释器加速。
• Full existing matrix: all eight V8 cases screened on three versions; four complete with three valid runs/version, four hit the 90-second timeout on every version. All five default microbench cases complete three runs/version. / 完整保留有效分数、超时和回退结果。
• Validation: 3018 Rust tests pass, 690 isolation canaries rejected, strict Clippy in three configurations, formatting/source-layout/Rust-only pass. 102037 Test262 result rows are identical before/after. Both frozen full-file checksum gates still exit 1 because only the source-fingerprint header changes; no frozen baseline changed. / Test262 逐条无变化；冻结文件校验的既有失败明确保留。

Bilingual report, measurements and raw-archive receipt / 双语报告与原始证据

Latest original PR CI failed at the inherited CLI integration-target inventory check (test262-host-metadata.py); the CLI targets and this check are unchanged by this PR. This is separate from the manual validation reported above.