[mstomar125]

Summary

Closes #2434.

The relationship picker used to have a paste-a-UUID entry path (relationship-viewer-modal.svelte) that bypassed the "already linked" filter search results went through, so pasting the id of an already-linked node offered a live "Use ID" Add whose write silently no-opped (the daemon's create_relationship is correctly idempotent for an existing edge).

Note on scope drift: my first pass at this fix (PR #2456, since closed) targeted the old stacked-table picker. #2432's two-pane redesign merged into main while that PR was in review and rewrote relationship-viewer-modal.svelte end to end, including removing the paste-a-UUID path entirely — the picker is search-only now. So the specific paste-UUID repro in #2434 can no longer happen. This PR instead:

1. Adds linkedTargetIds / isTargetLinked / filterUnlinkedTargets to relationship-grouping.ts as the single source of truth for "already linked into this group" — comparing case-insensitively (a caller may hold an id in a different case than the backend's canonical lowercase form; caught this in review of the closed PR).
2. runSearch in the two-pane picker now filters through filterUnlinkedTargets instead of an inline, unshared Set.
3. Fixes a real imprecision already present in the merged two-pane picker: the "no matches" message was keyed off group.rows.length > 0 (does this relationship have ANY linked row at all), so a genuinely-empty search on a relationship that already had other linked rows misreported "every result is already linked" instead of "No matches." It's now keyed off whether this search's own raw results were non-empty before filtering (addAllLinked).

The daemon's idempotent create_relationship is untouched — pure frontend diff.

Test plan

• 7 unit tests for linkedTargetIds / isTargetLinked / filterUnlinkedTargets (including 2 case-insensitivity regression tests), each individually mutation-tested (broke the real implementation, confirmed the test failed, restored, confirmed it passed again)
• bun run test — 232 files / 4895 tests pass
• svelte-check / tsc --noEmit / eslint — clean
• cargo build --bin nodespaced (the actual pre-push gate step) — clean
• Full pre-push gate (test:all + cargo build --bin nodespaced + test:e2e) — green, confirmed via the push itself

Note: bun run quality:fix's cargo clippy step currently fails on main due to unrelated code in packages/daemon/src/services/embeddings_service.rs (introduced by the most recent commit on main, unrelated to relationships). Confirmed this PR's diff touches zero Rust files and cargo build --bin nodespaced (what the actual push gate runs) is clean.

🤖 Generated with Claude Code