Dedupe redundant per-tick ContentIdle reads + gate the outer task-list fetch on DB change

[anutron]

Two follow-up CPU optimizations to the TUI's 1s onTick, companion to #944 (ce74221f).

1. agent.ContentIdle reuses detectNeedsInputSticky's already-computed signal

ContentIdle (BUG-036) ran right after detectNeedsInputSticky every tick but independently re-read each session's tail and re-ran ContentIdleFingerprint/ParkedSelectionSignal VT emulation — even though detectNeedsInputSticky's resumed-activity and content-stability passes had just computed the identical values for the identical tail moments earlier in the same tick.

ContentIdle gained a trailing cachedSignal func(id string) (ContentIdleSignal, bool) parameter: a cache hit skips the tail read + VT emulation entirely; ok=false (including a nil cachedSignal) falls back to the exact pre-existing independent computation, byte-for-byte. internal/api/push.go's daemon-side idle watcher (which has no equivalent shared per-tick pass) passes nil and is untouched. The TUI wires App.contentIdleSignalOf, sourced from a.needsInputRawSignals (the dirty-check cache from #944), which now also threads through the fp that pass was already computing and discarding.

Safe because it's the literal same function call, not an approximation — ScreenRenderer.render's RIS-reset-per-call idempotency (already relied on by #944) guarantees identical output for identical (tail, cols, rows).

2. refreshTasksWithIDs's outer a.db.Tasks() fetch is gated on DB change

Mirrors hera.HeraPage's existing shouldRebuild/markRebuilt/InvalidateChangeGate PRAGMA data_version pattern, applied to the plain task-list fetch — but only on the periodic onTick path (new allowCachedTasks param). Every other caller (refreshTasksLocal, refreshTasksAsync, a direct bulk-prune call) keeps unconditional-fetch, since db.DB.DataVersion's documented same-connection blind spot means the gate can never be trusted to see a write made by the very call that would also be consulting it.

A code review of the first pass at this caught that it only handled the one write site named in scope (reconciliation's own SetStatus) and missed four more: several hera mutations funneled through heraRefresh() (nuke/cascade-nuke/step-status), startSession (called from 5 sites, only 2 of which separately refresh), switchSession's live-session branch, and the async session-ID recapture path in handleSessionExitUI. All five now call invalidateTasksChangeGate() at their write site; each has a dedicated regression test that fails when its invalidate call is removed (verified by hand before writing this up).

One accepted, named, bounded gap remains: internal/agent/autorename.go's background rename goroutine has no App reference to invalidate through — cosmetic-only (a suggested rename lags until an unrelated write), documented in context/knowledge/gotchas/events.md.

Testing

• New differential/equivalence tests for both fixes (TestContentIdle_CachedSignal, TestRefreshTasksWithIDs_GatedFetch, plus 4 regression tests pinning each newly-found write site) — each regression test verified to fail when its fix is reverted.
• make pre-pr green except make vuln (pre-existing stdlib CVEs, continue-on-error in CI — not introduced by this PR).
• Full -race suite green; coverage 88.7% (floor 88).

Specs: No behavioral changes (internal perf/caching only, no observable behavior change) — no openspec/ change folder, matching precedent set by #944.

🤖 Generated with Claude Code

[github-actions]

Merging this branch will increase overall coverage

Impacted Packages	Coverage Δ	🤖
github.com/drn/argus/internal/agent	89.94% (+0.04%)	👍
github.com/drn/argus/internal/api	87.00% (ø)
github.com/drn/argus/internal/tui	83.27% (+0.04%)	👍

---

Coverage by file

Changed files (no unit tests)

Changed File	Coverage Δ	Total	Covered	Missed	🤖
github.com/drn/argus/internal/agent/needsinput.go	95.24% (+0.14%)	273 (+8)	260 (+8)	13	👍
github.com/drn/argus/internal/api/push.go	94.88% (ø)	254	241	13
github.com/drn/argus/internal/tui/app.go	78.92% (+0.22%)	2785 (+33)	2198 (+32)	587 (+1)	👍
github.com/drn/argus/internal/tui/heraactions.go	76.21% (+0.04%)	580 (+1)	442 (+1)	138	👍

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/agent/needsinput_test.go
• github.com/drn/argus/internal/api/push_test.go
• github.com/drn/argus/internal/tui/app_test.go
• github.com/drn/argus/internal/tui/smoke_test.go