Reduce resolver allocation and bound candidate loops

[jdalton]

Tooling update

Rebased onto merged #212 as one commit. Sources and tests use .mts, Node tests use Vitest, and the published JavaScript paths stay unchanged. The original engine changes are preserved.

Local lint, formatting, Node and packed-package checks pass. All 41 WPT pages pass against both generated builds.

Change

Extract the remaining resolver-execution work from #167: bounded candidate loops, direct cached-plan fetches, null no-op resolvers, raw match-resolver arrays, captured slicing, and early tag rejection within a compound selector.

This applies to master independently. It does not include #188's context-free plan caching, the attribute-equality optimization, or legacy-DOM handling. Those changes have separate destinations and must be reconciled when combined.

The extraction corrects two archived item-loop defects: the result index was uninitialized, and the tail replaced the result array with true. Compiled-cache keys now distinguish mode and callback presence, preventing a cached array loop or no-op from being reused for a different contract. Installed wrappers are tested with the captured slicing helper.

Validation: all 7 focused Node tests and 4 shared-runtime tests pass with Node 26.5.0 and jsdom 30.0.1. Coverage includes cold/cached queries, mutation, callbacks, escaped classes, cache turnover, both collection modes, tag-read ordering, and installation. Commands and scope are in docs/resolver-execution.md.

No new performance claim is made; representative benchmarks should be rerun on the combined branch.

[jdalton]

Integrated the merged legacy DOM and nth-index fixes. Added Node and WPT tests for bounded array/item loops and for adding, replacing, and omitting callbacks after matching is cached. Matching-cache keys now distinguish callback presence. All 465 Node tests and 49 WPT pages pass; CI is green, and coverage thresholds are unchanged.