Add uninitialized reduction and indexed replay

[shinaoka]

Summary

• add full-overwrite execute_uninit replay for erased reduction, gather, dynamic slice, dynamic update, and scatter plans
• keep copy-then-update state private through a non-escaping initialized logical-layout receipt, without constructing initialized backing slices over holes
• preserve explicit execution contexts, parallel thresholds, reduction association, wrapping integer scatter, and expected-serial additive scatter
• add typed-storage differential, lifecycle, hole, error, wrapping, parallel, Miri, source-contract, and performance coverage

Verification

• cargo test --workspace
• focused default/parallel: indexed 68, reduction 11, source contracts 5
• cargo check --workspace --all-targets
• cargo check -p strided-kernel --all-targets --all-features
• RUSTDOCFLAGS="-D warnings" cargo doc --workspace --no-deps
• strict-provenance + symbolic-alignment Miri focused lifecycle/hole/wrapping cases
• coverage: 53/53 files passed
• Sol low/high final review: no findings

Performance gate

Candidate-relative upper 95% ratios, 31 alternating pairs, affinity isolated:

row	t1 CPU60	t4 CPUs60-63
reduce	1.0169	1.0456
gather	1.0415	1.0337
dynamic slice	1.1028	0.9308
dynamic update	1.1966	1.1312
scatter serial	0.7869	0.7856

All rows pass the 1.20 gate.

Closes #187