feat(output)!: mandatory/essential columns and explicit --fields opt-out

[jpage-godaddy]

Summary

• Add TableColumn::essential(bool): a column marked essential is never hidden or truncated for terminal width, even if it alone exceeds the display width (still capped at NO_TRUNCATE_MAX_WIDTH for pathological values). Use it for fields a view is useless without — e.g. a DNS record list needs type/name/data at minimum.
• An explicit --fields selection (as opposed to a command's default_fields fallback) now disables width-based hiding and truncation entirely for the columns it selects, regardless of their essential flag — the user named exactly what they want to see, so the row simply overflows the terminal if it must. Hiding/truncation remain a convenience reserved for the default view, where cli-engine (not the user) picked the field set.
• Reworked the column-dropping algorithm in output/human/{columns,body}.rs from a contiguous-prefix model to an index-set model so essential columns can be exempted regardless of their declared priority position.
• Breaking: render_human_with_view, render_human_with_registry_selected, and the two now-dead registry wrapper functions (render_human_with_registry, render_human_with_registry_for_schema, deleted outright) are now crate-internal (pub(crate)). They had no production callers beyond middleware's single internal use — their only other callers were command-module tests reaching past the engine to render a view against fixture data, which is exactly how essential/fields_explicit ended up breaking those call sites in the first place.
• Added preview_human_view(data, columns) as the one stable, intentionally minimal public entry point for that test use case: render a TableColumn view against fixture data the way a default view would, no Envelope or CLI run required. It's deliberately decoupled from the internal renderers so future rendering-behavior changes won't need to break it the way this one did. Migrated cli-engine's own tests and the downstream gddy consumer's ~12 call sites to it; moved the handful of tests that genuinely exercise internal registry/custom-renderer dispatch contracts into crate-internal unit tests instead.
• Split output/human/tests.rs into a tests/ directory module (grouped by concern: footer, alignment, field_selection, width_fitting, nested, value_format, registry) since it grew past the 1000-line module cap.
• Updated docs/concepts.md to document the essential/explicit-fields behavior and the new preview_human_view testing entry point.

Test plan

• cargo fmt --all --check
• cargo clippy --all-targets -- -D warnings (default and --features pkce-auth)
• cargo test --all-targets (293 passed; 338 with --features pkce-auth)
• cargo test --doc (default and --features pkce-auth)
• RUSTDOCFLAGS='-D warnings' cargo doc --no-deps (default and --features pkce-auth)
• cargo rustdoc --lib -- -W missing-docs (zero missing docs)
• ./cli-engine/scripts/check-module-size.sh
• All of the above re-verified after rebasing onto main (rustc 1.99.0, as pinned by rust-toolchain.toml)
• Manually verified against a real consumer command (gddy dns list) in a sibling worktree: 17-column view with type/name/data essential correctly hides 7 non-essential columns at 80-column width while the three essential ones always survive
• Migrated and re-tested the gddy consumer's ~12 render_human_with_view call sites against preview_human_view — all 991 of its tests pass

BREAKING CHANGE: render_human_with_view and render_human_with_registry_selected are now crate-internal (no longer exported from the crate root or output); render_human_with_registry and render_human_with_registry_for_schema are removed entirely. Any consumer calling these directly (expected mainly in command-module tests) should switch to the new preview_human_view(data, columns).

🤖 Generated with Claude Code

[Copilot]

Copilot review overview

🟡 Changes recommended

Two exported renderer signatures are source-breaking, and the row-selection implementation unnecessarily duplicates retained payload data.

Review effort: Balanced
Findings: 2 · 1

Open (3)

• Cloning retained cells duplicates large response payloads · New
• Adding parameter breaks exported function callers · New
• Explicit selection still caps cell width · New

What changed in this PR

Adds essential human-output columns and honors explicit --fields selections without terminal-width fitting.

Changes:

• Adds TableColumn::essential and index-based column fitting.
• Propagates explicit field-selection state through rendering.
• Splits and expands human-output tests and documentation.

File	Description
cli-engine/​tests/​foundation.rs	Updates renderer calls.
cli-engine/​tests/​exhaustive_output.rs	Updates renderer calls.
cli-engine/​tests/​custom_human_views.rs	Updates registry rendering call.
cli-engine/​src/​output/​human/​tests/​width_fitting.rs	Tests width and essential-column behavior.
cli-engine/​src/​output/​human/​tests/​value_format.rs	Houses value-format tests.
cli-engine/​src/​output/​human/​tests/​nested.rs	Houses nested-output tests.
cli-engine/​src/​output/​human/​tests/​mod.rs	Declares split test modules.
cli-engine/​src/​output/​human/​tests/​footer.rs	Houses footer tests.
cli-engine/​src/​output/​human/​tests/​field_selection.rs	Tests field ordering and selection.
cli-engine/​src/​output/​human/​tests/​alignment.rs	Houses alignment tests.
cli-engine/​src/​output/​human/​tests.rs	Removes the monolithic test module.
cli-engine/​src/​output/​human/​mod.rs	Adds essential columns and explicit-selection APIs.
cli-engine/​src/​output/​human/​columns.rs	Implements index-based width fitting.
cli-engine/​src/​output/​human/​body.rs	Applies essential and explicit-field behavior.
cli-engine/​src/​middleware/​run.rs	Passes explicit-selection state to rendering.
cli-engine/​docs/​concepts.md	Documents width-fitting semantics.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🔵 Needs a closer look

The rendering logic appears coherent and well tested, but the intentional removal of exported renderer APIs warrants final human review for downstream compatibility.

Review effort: Balanced
Findings: None

Resolved since last review (3)

• Adding parameter breaks exported function callers
• Cloning retained cells duplicates large response payloads
• Explicit selection still caps cell width

[Copilot]

Copilot review overview

🟢 Approval recommended

The rendering behavior, middleware plumbing, documentation, and focused regression coverage are consistent with the stated design.

Review effort: Balanced
Findings: None

[mguerrero3-godaddy]

I see this has merge conflicts. Let me know If this get stalled and needs reapproval.