assay-tui S03 review backlog: doc/comment/minor code improvements

[wollax]

Backlog items surfaced during the second PR review of #162 (S03 Chunk Detail View and Spec Browser). The critical and structural issues were fixed inline; these are doc, comment, and minor code quality improvements.

Items

detail_spec_note field doc (#2)

The doc says "Diagnostic reason when detail_spec is None" but the field is also written on history load failure when detail_spec is Some. Update to reflect both uses, e.g.:

General-purpose diagnostic message for the chunk detail view. Set when either spec loading or history loading fails. Rendered as the block title warning when spec is present; rendered as the fallback paragraph when spec is absent.

join_results doc missing run = None case (#3)

The most common real-world case (run is None because no gate history exists yet) is not documented. Add as the first bullet:

run is None (no gate history) → all criteria return None (Pending)

ChunkDetail Esc arm lacks preservation comment (#4)

detail_milestone and detail_list_state are intentionally not cleared on Esc (D099 — cursor position preserved). Without a comment a future maintainer could break this. Add:

// Intentionally do NOT clear detail_milestone or detail_list_state:
// returning preserves the user's chunk cursor position (D099).

draw_chunk_detail doc only describes happy path (#5)

Current: "Render the chunk detail screen with a table of criteria and their results."
The function has two modes. Update to describe both the table path (spec=Some) and the paragraph fallback (spec=None).

.last().unwrap() safety annotation (#6)

Ok(ids) if !ids.is_empty() => {
    let run_id = ids.last().unwrap().clone();

Use .expect("non-empty: guarded by !ids.is_empty() above") to make the invariant explicit at the call site.

up_down_in_milestone_detail weak assertion (#7)

assert!(after_down > initial_selection) passes for any forward move. Pin to concrete indices:

assert_eq!(app.detail_list_state.selected(), Some(1), "Down from 0 must select 1");
assert_eq!(app.detail_list_state.selected(), Some(0), "Up must wrap back to 0");

Validation comment in setup_project incomplete (#8)

// Validation requires at least one criterion with a `cmd` or `path` field.

Missing third accepted form. Fix:

// Validation requires at least one criterion with a `cmd` field, `path` field,
// or `kind = "AgentReport"` — at least one criterion must be executable.

Error vs. info styling in draw_chunk_detail (suggestion)

"Legacy flat spec…" (informational) and "Failed to load spec…" (error) both render .dim(). Errors should use .fg(Color::Red); info should stay dim. Could detect by prefix or add a separate detail_spec_error: Option<String> field.

detail_run unnecessary clone (suggestion)

let run_id = ids.last().unwrap().clone();
match history::load(&assay_dir, &chunk_slug, &run_id) {

history::load accepts &str; ids.last().unwrap() (a &String) already coerces. Drop the .clone() and the binding.

---

Discovered during second parallel review of PR #162. All critical and structural issues already addressed; these are low-risk doc/comment/minor code cleanups safe to batch into S05 or a dedicated polish pass.