fix: make retired touch observations consistent

[thymikee]

Summary

Selector and coordinate touch response construction now discard settle observations and reference frames when their admitted session lifetime has ended. Both branches read the same current lifetime after any frame probe. Mandatory action finalization still refuses writes to a successor session.

Removed unused captured-record arguments from selector and find capture and touch response construction. The blocked-directory open control now asserts the actual ENOTDIR error after native dispatch.

Nine files changed. Part of #3116, stacked on #3150; addresses the non-point touch finding on #3148 and optional cleanup on #3145.

Validation

Validated a4b723e799. The new selector retirement control fails against the pre-fix code (1 failed | 27 passed); all 44 focused controls pass after the fix. Quick and Fallow checks pass. Independent review found no actionable findings.

pnpm check:affected --base refactor/session-action-lifetimes --run passes selected runnable gates and 1,050 related tests across 180 files. CI and the requested live device evidence remain pending.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	4.96 MB	4.96 MB	-51 B
Package (unpacked)	4.96 MB	4.96 MB	-51 B
Package (download)	1.49 MB	1.49 MB	-7 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	18.4 ms	19.2 ms	+0.8 ms
CLI --help	55.0 ms	54.1 ms	-0.9 ms

[cubic-dev-ai]

All reported issues were addressed across 9 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[thymikee]

This PR is ready at a4b723e. All 14 checks pass on this head, and I know of no conflicts. Not blocking, and you can take or leave it: buildFillResponsePayloads at https://github.com/callstack/agent-device/blob/a4b723e/src/daemon/interaction/internal/interaction-touch-fill.ts#L213 still calls sessionStore.requireCurrent after dispatch, so a lifetime that retires during a fill would throw session_lifetime_ended instead of dropping the reference frame and settle as the touch path now does. A follow-up could use resolveCurrent there, as buildTargetedTouchResponsePayloads does, with a matching retired=true control.

I did not run the focused tests or the pre-fix red run, so regression validity comes from reading the base code path. I did not review live device evidence, because the live-session route is unchanged and the new unit control covers the retired path. That control calls buildTargetedTouchResponsePayloads directly, but the press caller passes params straight through, so the bypass is thin. Nothing else is needed before merge, and the PR then lands after the stacked base #3150.

The Cubic P3 thread still applies and is optional: #3151 (comment). The retired point touch still runs the reference-frame probe and logs a warning, but the base commit behaves the same, so this is noise and not a regression.

[thymikee]

Consolidated into #3144 as part of reducing #3116 to seven PRs. The composition preserves the complete pre-consolidation source tree, including tests and later review corrections. This PR is superseded; its review discussion and native evidence remain available. Outstanding findings transfer to the owning keeper in the implementation record.

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-10-03 21:17 UTC