fix(tui): bulk-edit include toggles invisible + raw custom-field labels (v1.6.1)

[thomaschristory]

Closes #137. Patch release v1.6.1.

Problem

In the TUI bulk-edit form, custom-field edits silently didn't save and showed an ugly custom_fields.<name> name.

Root causes & fixes

1. Invisible include toggle → silent no-save (the real bug). .bulk-include { width: 6 } is too narrow for a Textual Switch (border + padding + slider ≈ 10 cells). With the real stylesheet the slider clipped to zero width, so every per-field include toggle rendered blank. No visible toggle → no field opted in → empty patch → silent no-change. Fixed by sizing .bulk-include to fit (width: auto), with a regression test asserting the slider never clips to zero.
2. Raw custom_fields.<name> labels. Both forms rendered Label(name) instead of CustomFieldDef.label. Now WidgetSpec carries label; the form rows and the change-preview diff show the human label.
3. Latent destructive bug. A select whose choice-set didn't resolve rendered empty; opting it in seeded None and silently nulled the field. A blank select now seeds nothing on opt-in; only the explicit ∅ button nulls.

Tests

7 new tests (TDD, each watched fail first): toggle-not-clipped regression (real CSS loaded), human-label rendering in both forms + diff, end-to-end custom-field save, and blank-select-doesn't-null.

Docs

• CHANGELOG.md — v1.6.1 section.
• docs/guides/interactive-tui.md — clarified the include-toggle / per-field opt-in and that custom fields appear under their label.

Release

Version bumped to 1.6.1 across pyproject.toml, nsc/_version.py, uv.lock, and CHANGELOG. Tag v1.6.1 to be pushed on main HEAD after merge.

🤖 Generated with Claude Code

[thomaschristory]

Claimed by maintainer's agent. Branch release/1.6.1. Running adversarial review next, then merge + tag v1.6.1.

[thomaschristory]

Adversarial review (3 parallel read-only agents) + fixes

Correctness/regression, break-the-fix, and test-quality reviews ran against the diff. Verdicts: ship, with findings.

Findings actioned:

• Data-loss bug, newly reachable — because the toggle is now usable, opting in a custom field whose widget didn't reflect the records' shared value would overwrite it: a shared boolean rendered False and opting it in unchanged flipped all records to False; a shared multiselect cleared. Fixed: custom-field widgets (incl. multiselect) now seed the shared value over flattened custom_fields.<name> keys → opt-in-without-edit is a no-op. (+2 tests)
• on_select_changed could stage a mount-time/blank NULL — added a guard so a blank select can't stage a null before opt-in. (defensive)
• Test gap — added a single-edit diff-modal label assertion (live wiring path).

Findings dismissed (with reason):

• detail.py diff_rows missing field_labels — detail view edits only top-level body fields (no .label); custom fields there are non-editable flattened rows, so it'd be dead code.
• bulk tags don't seed shared — pre-existing intentional design (tags have their own {name,slug} flow); left untouched.

Full suite: 1335 passed, 1 skipped; ruff + mypy --strict clean.