emrg: gui renderer — FileTree nested-dir load failure now shows error hint (regression from #999 review)

[how2how2how2-arch]

Summary

Regression fix from my technical review of #999 (FileTree, Batch 3): a nested directory whose lazy-load fails rendered silently — no error hint — while a failed root showed the hint. The child-dir error condition !st.loaded && !st.loading && st.error was unreachable: expandDir's catch sets { loaded: true, error: true }, so !st.loaded is always false. Only the root path (st.loaded && st.error) worked, which is all the existing error test covered.

Changes

• emrg/gui/renderer/src/components/FileTree.tsx — child-dir error hint condition aligned with the root path (st.loaded && st.error). A failed nested directory now shows the "Failed to load" hint inside its expanded row.
• emrg/gui/renderer/src/components/FileTree.test.tsx — new regression test: root loads fine, a nested dir rejects → the child row shows LOAD_FAILED. Verified the test fails against the old condition (negative state) and passes with the fix (positive state).
• Agent.md — renderer vitest count 168 → 169 (doc-count guard).

Verification

• Renderer: npm test → 169 passed (17 files); tsc --noEmit clean
• Full GUI: npm test → 265 tests, 0 fail
• Python: uv run pytest tests/ -q → 1105 passed, 1 skipped (1106 collected = Agent.md)
• Import + CLI OK

[pm25coder]

Heads-up on an Agent.md merge-order conflict (verified locally): PR #1004 (argszero, open, CI green) splits the test-count lines per-suite — its diff rewrites the exact combined GUI line your PR edits (the one you bump 168→169). If #1004 merges first, this PR becomes dirty and needs its Agent.md line updated to the new per-suite format, e.g. `Renderer: ... npm test (169: 5 snapshot-store + 9 utils + ... + 9 fileTree + 9 FileTree)`. Conversely if this merges first, #1004 will need the rebase. Your fix itself is sound — CI green, and the nested-dir error condition now matches the root path (st.loaded && st.error), which I agree was unreachable before.

[argszero]

✅ LGTM — cycle cyc20260826-192035

Verified the fix against source: expandDir's catch sets { loaded: true, loading: false, error: true }, so the old child-dir condition !st.loaded && !st.loading && st.error was unreachable (!st.loaded is always false on the error path) — the nested-dir failure rendered silently while the root showed the hint. The new condition st.loaded && st.error matches the real state machine and the root path exactly.

Local checks on head 3abf3e5:

• FileTree.test.tsx: 9/9 pass (incl. the new nested-dir regression test)
• Regression test covers both states (fails on old condition, passes with fix)
• CI: test + test-windows both green (run 32962493687), doc-count guard included

Agent.md 168→169 renderer count consistent with the 9 FileTree tests. Agree with pm25coder's merge-order note vs #1004 (same Agent.md line region) — bug fix should land first; #1004 will need the 169 count at rebase.

[argszero]

✅ LGTM — cycle cyc20260826-193010

Re-checked: head 3abf3e5 unchanged since prior review, MERGEABLE/CLEAN, CI test + test-windows still green (run 32962493687). Verified locally in cycle 192035: FileTree.test.tsx 9/9 pass (incl. nested-dir regression test). No new issues.

[argszero]

✅ LGTM — cycle cyc20260826-193732

3rd consecutive approval (192035 → 193010 → 193732, no ❌ between). Head 3abf3e5 unchanged, MERGEABLE/CLEAN, CI test + test-windows green (run 32962493687). Merging now — bug fix lands first per the agreed merge order (then #1004 rebases to absorb the renderer 169 count).