Fix ctrl+g/ctrl+b excursion re-arm to track needs-input identity, not count (BUG-069)

[anutron]

Summary

• The ctrl+g/ctrl+b "problem-child excursion" (Rework ctrl+g into a problem-child excursion with ctrl+b manual restore #906/Tighten ctrl+g/ctrl+b help-overlay wording #908) re-arm path (excursion == nil after a restore, needs-input count still >=1) froze a fresh snapshot on the very next rebuild regardless of whether anything genuinely new had happened. A stale, never-resolved needs-input role (e.g. an unrelated stuck (?)) kept the count >=1 indefinitely, so every ctrl+b re-armed and froze immediately — silently discarding any fold/selection change the operator made afterward until the next explicit restore.
• Confirmed against two back-to-back live dogfood repros (verbatim in the originating task): ctrl+b replayed a stale, much-earlier layout instead of the operator's actual prior position, and in the second (clean, controlled) test replayed the FIRST test's world instead of the second test's own manual navigation.
• Fix: Rail.armedNeedsInputIDs (Model.needsInputRoleIDs, hasNewNeedsInputID) tracks the needs-input set by role identity, not count. While no excursion is held, the baseline keeps refreshing to the current set on every rebuild for as long as it stays a subset of the last capture, and only freezes the instant a genuinely new, distinct role id appears — so the eventual freeze reflects the operator's latest position, never a stale one. The baseline updates unconditionally even while an excursion is frozen, so a role that merely folds into an already-open excursion doesn't look "new" again after a later discharge (would otherwise spuriously re-freeze).
• Separately confirmed (not fixed — narrow, pre-existing, shared with BUG-002's persisted-selection mechanism): currentRef() has no identity encoding for a cursor resting on an Archive-expando or Freelance-header fold row, so restoreCursor silently no-ops in that specific case. Documented as a known follow-up candidate, not fixed here since the clean repro is fully explained by the identity-tracking bug alone.
• Updated openspec/specs/hera-view/spec.md's excursion requirement/scenarios (same-feature bug fix, no new change folder per project convention) and context/knowledge/gotchas/hera-view.md to correct the previously-wrong "intentional, not a bug" description of the count-edge re-arm.

Test plan

• New regression tests in internal/tui/hera/excursion_test.go: TestRail_ExcursionSnapshot_StaleItemDoesNotFreezeAwayManualNavigation (fails against pre-fix code, confirmed live), _DoesNotReArmAfterRestoreWithoutNewEntrant, _EntrantAbsorbedWhileFrozenDoesNotReFreezeAfterDischarge, TestModel_NeedsInputRoleIDs, TestRail_CurrentRef_ZeroOnArchiveExpandoRow (documents the separate confirmed-but-unfixed gap)
• go build ./..., go vet ./..., make fmt-check, make lint-pr (golangci-lint v2.11.4, 0 new issues)
• make test-cover-gate (race suite + 88% coverage floor) — green at 88.7%
• openspec validate hera-view --strict — valid
• make vuln reports 3 pre-existing stdlib CVEs (GO-2026-5856/5039/5037, unrelated to this change) — CI runs this step with continue-on-error: true per .github/workflows/ci.yml

🤖 Generated with Claude Code

Co-Authored-By: Claude noreply@anthropic.com

[github-actions]

Merging this branch changes the coverage (1 decrease, 1 increase)

Impacted Packages	Coverage Δ	🤖
github.com/drn/argus/internal/tui	82.69% (-0.03%)	👎
github.com/drn/argus/internal/tui/hera	92.12% (+0.07%)	👍

---

Coverage by file

Changed files (no unit tests)

Changed File	Coverage Δ	Total	Covered	Missed	🤖
github.com/drn/argus/internal/tui/hera/model.go	96.52% (+0.13%)	345 (+12)	333 (+12)	12	👍
github.com/drn/argus/internal/tui/hera/rail.go	92.56% (+0.07%)	874 (+8)	809 (+8)	65	👍

Please note that the "Total", "Covered", and "Missed" counts above refer to code statements instead of lines of code. The value in brackets refers to the test coverage of that file in the old version of the code.

Changed unit test files

• github.com/drn/argus/internal/tui/ctrlg_test.go
• github.com/drn/argus/internal/tui/hera/excursion_test.go

[anutron]

Update: found and fixed a second, related bug (BUG-070) while dogfood-testing the BUG-069 fix above.

Live repro: relaunching the TUI while a stale, never-resolved needs-input role already existed (predating the launch) caused ctrl+g → ctrl+b to leave both panes empty and flip unrelated orchestrators' fold state to a stale prior-session layout, discarding the operator's actual navigation.

Root cause: Rail.noteExcursionTransition's fresh-interruption arm fired unconditionally on the rail's very first-ever SetModel call — before buildRows had run even once. At that instant currentRef() has no rows to resolve (→ selRef==0, a silent no-op on restore) and the fold-state snapshot reflects whatever the previous session's SetStateStore load persisted, not anything from the current session.

Fix: noteExcursionTransition now early-returns (seeding the needs-input baseline only, no capture) on the literal first-ever rebuild. Normal arm/re-arm rules resume from the second rebuild on; ctrl+g's own EnsureExcursionArmed still arms from the operator's live position if nothing new interrupts them before their first press.

Added regression tests (TestRail_ExcursionSnapshot_ColdStartDoesNotCaptureBogusSnapshot / _ColdStartWithNoPriorNeedsInputArmsNormally), updated openspec/specs/hera-view/spec.md and context/knowledge/gotchas/hera-view.md. make pre-pr clean. Confirmed fixed live via dogfood re-test (relaunch → repeat the ctrl+g/ctrl+b sequence → restores to the actual post-navigation position). CI green on the new commit.