[zkochan]

Native add now owns selectors/ignores, identity and scope resolution, main-file and root ownership rules, package/source links, cache publication and bitmap/history persistence. This includes namespace variants, scoped vendor components, unchanged re-add history, partial-link failures and published best-effort backup/temporary-cleanup behavior.

Published Harmony providers remain adapters for environment resolution, semantic docs, component-load hooks and package transforms. Unsupported grammar/configuration/filesystem layouts delegate before effects; failures after provider or publication effects remain owned. Successful CI adds stay native; CI error cases requiring the released JavaScript stack delegate before effects. No Node CLI sources or third-party crate versions change.

Validation against Node 22.22.0 / Bit 2.2.94:

• All seven required checks pass: formatting, Clippy, Dylint, rustdoc, 1,373 workspace tests (one existing ignored test), 592 e2e tests and port-plan inventory. Focused export package/callback checks also pass.
• Strict paired commands use the same absolute path and compare exact output/status, workspace/repository files, modes, symlinks, checksum-validated caches, bitmap history and real file-remote state. Custom providers are physically compiled and exercised through their actual lifecycle.
• Release qualification at measured commit e3ae55f: 90 add pairs (80 native, 10 explicitly delegated CI errors) plus 45 status/checkout/export regression pairs. All parity checks pass; every native median is faster in non-TTY, TTY and CI modes. Non-TTY: single 269 vs 441 ms; 100 components/1,100 files 1,162 vs 1,565 ms; root 280 vs 353 ms; re-add 275 vs 442 ms (Rust vs Node).
• Peak aggregate CPU load was 16.5% after a quiet-minute gate. A representative contended run is retained as invalid timing evidence. These are warm builtin-metadata/file-remote fixtures; no first-use, network or cross-platform speed claim. Raw samples, fingerprints and reproduction scripts are included. The final outside-workspace diagnostic, filesystem-aware main-file casing and exact platform-specific test expectation changes passed another 135 production-source parity pairs before the test-only fixture-ordering fix; their 92.3%-peak contended timing records are explicitly invalid and excluded from speed claims.
• Callback-failure oracles retain exact output/state comparisons using one component, avoiding the released concurrent loader’s nondeterministic batch order. Separate two-component linker tests verify the transformer barrier and partial-write prefix. Production code is unchanged by this fixture fix.
• Windows CI gets a dedicated tracking parity group. Platform-native diagnostic/package paths, blocked-parent errors and filesystem-aware explicit main-file lookup and strict inferred/unresolved casing on all platforms are covered without path masking. Offline doctor catalog tests use a genuine CONNECT/TLS response with a fixture-only trusted certificate, preserving exact diagnosis and request counts. Original ADD inventory coverage is 35/79 cases; clone/lane and unrelated mutation cases remain unchecked.

Part of #73. The remaining tracking/mutation roadmap stays open. Base is rust-workspace-materialization, so merging this PR feeds upstream teambit#10813; that upstream PR remains open.