Improve extension architecture helpers

[Kingson4Wu]

Harness assessment

• WorkOrder: 1788201000000-fluent-frame-harness-auto
• Base: dev
• Strategy: health-first, bounded to confirmed actionable risk or quality gaps.
• Initial and final architecture assessment score: 93/95.
• The assessment produced generic guarded architecture slice prompts; no confirmed production bug or actionable security finding was found.

Selected subtasks

• Architecture slice 1: shared extension runtime reload error classification between content and learning-generation messaging paths.
• Architecture slice 2: shared popup relative-time formatting while preserving queue minute-only copy and library hour-capable copy.

Skipped or no-op subtasks

• Bug-fix: skipped because no confirmed production-risk bug was proven in the checked extension/native/shared/persistence/scheduling surfaces.
• Security-maintenance: skipped after pnpm audit --audit-level high reported no known vulnerabilities and no reachable high-signal security issue was confirmed.
• Test-coverage: skipped because the workspace has no reliable unified coverage command, and no meaningful uncovered high-risk behavior was identified beyond the focused regression tests added with the architecture slices.
• Architecture slice 3: attempted YouTube URL candidate helper extraction was rejected by the repo's content-script bundle architecture guard, so it was fully reverted and treated as a no-op.

Changes

• Added apps/extension/src/chromeRuntimeErrors.ts and regression tests for reload-context error detection.
• Updated content.ts and learningGenerationClient.ts to use the shared runtime error helper without changing user-facing error messages.
• Added apps/extension/src/popupRelativeTime.ts and regression tests for invalid, future, seconds, minutes, hours, and queue minute-only behavior.
• Updated popupQueue.ts and popupLibrary.ts to use the shared popup formatter while preserving existing displayed copy.

Verification

• pnpm run typecheck passed.
• pnpm test passed: shared 42 tests, extension 150 tests, native-host 132 tests, scripts 48 tests.
• pnpm audit --audit-level high passed with no known vulnerabilities.
• Final assessment command completed and still reported score 93/95.
• Worktree is clean on loop/fluent-frame/harness-auto/1788201000000-fluent-frame-harness-auto.

Remaining risk

• The target score 95 was not reached because the assessment script continues to emit generic guarded architecture slice prompts after the accepted slices.
• The third proven candidate was intentionally not retained because it violated the content-script bundle boundary.
• Auto-merge should only proceed after GitHub mergeability and CI checks pass.