fix(tv-motion): route the pre-#276 scroll-follow sites through the TV motion profile

[ghbarker]

Why

Reported: the TV Motion Profile setting (Smooth vs. Snappy, PR #276) has no noticeable effect on a real device — "smooth is still very snappy."

Root cause found (with evidence)

The Smooth/Snappy wiring itself is correct end to end and is not the bug:

• The settings row persists the choice via TvMotionController.select, which sets TvMotionController.notifier.value synchronously (lib/services/tv_motion_profile.dart).
• main.dart's _DebrifyAppState subscribes with an explicit addListener/setState (documented as deliberate, above a ProfileGate that only rekeys its child) and rebuilds TvMotionScope on every change.
• AppMotion.of(context) reads TvMotionScope.of(context) in build()/didChangeDependencies() at every call site checked (lib/theme/widgets/focus_expression.dart, lib/theme/widgets/hover_grow.dart, lib/widgets/tv_sidebar_nav.dart), so a profile switch retargets tweens on the next frame.
• This is already covered by passing tests: test/theme/tv_motion_profile_test.dart ("AppMotion.of a scope change rebuilds the dependents") and test/settings_tv_motion_row_test.dart ("a chip press rebuilds a real AppMotion.tvFocus consumer").
• The numeric gap is real and not accidentally aliased: snappy is fast (120ms) + standard curve, smooth is base (220ms) + emphasized (overshoot) curve — different durations and different curve shapes (lib/theme/app_motion.dart).

The actual bug: several DPAD scroll-follow sites bypass AppMotion entirely and hardcode a duration, so the profile switch never reaches them.

1. Three sites predate PR Add a TV motion profile: Smooth or Snappy, chosen per box and overridable #276 (which merged 2026-09-07 21:44) by roughly two days and were simply never migrated:

   • lib/screens/search/board_cell.dart:548 (_StremioCard, the Home/Discover board card) — hardcoded widget.isTelevision ? 140ms : 260ms, no profile branch at all.
   • lib/screens/search/favourite_art_cell.dart:329 (ArtPoster) — identical hardcoded 140/260ms pattern.
   • lib/widgets/episodes_panel.dart:2582 (_CompactEpisodeRow) — hardcoded 220ms, unlike its sibling EpisodeTile.onFocusChange which already read AppMotion.tvScroll correctly.

2. Three more were added by PR fix(detail): scroll-follow the rec rail/grid cursor in Marquee, Dossier, Console #281 ("fix(detail): scroll-follow the rec rail/grid cursor in Marquee, Dossier, Console"), which merged at 21:39 — six minutes before Add a TV motion profile: Smooth or Snappy, chosen per box and overridable #276 — so its branch point predates the motion profile:

   • lib/widgets/detail/detail_layout_console.dart:1031
   • lib/widgets/detail/detail_layout_dossier.dart:684
   • lib/widgets/detail/detail_layout_marquee.dart:450

   All three hardcoded duration: Duration.zero unconditionally for the rec grid/rail's scroll-follow-on-focus.

git merge-base --is-ancestor dfaea271 d0d9c5d6 confirms PR #281's merge commit is an ancestor of PR #276's — i.e. #281's code was written and merged against a tree that didn't have the motion profile yet, and nothing revisited it afterward.

Device detection (TvMotionProfile.defaultFor) and the settings row's persistence/read path were both reviewed and are correct — not the cause.

What

Routed all six sites through the shared profile, matching the pattern already used correctly elsewhere (catalog_item_tile.dart, detail_identity.dart, episode_tile.dart, iptv_channel_row.dart):

• board_cell.dart / favourite_art_cell.dart: duration: motion.scrollTempo(isTelevision, 260ms, tvSnappy: 140ms) — snappy keeps the shipped 140ms glide unchanged, smooth now reaches AppMotion.tvScroll (260ms).
• detail_layout_console/dossier/marquee.dart: duration: tv ? motion.tvScroll : Duration.zero — snappy stays an instant jump (unchanged), smooth now glides.
• episodes_panel.dart: duration: motion.tvScroll directly (this row is TV-only), same before/after contract as its EpisodeTile sibling.

Off-TV behavior is untouched everywhere.

How verified

• New regression tests that fail on the pre-fix code and pass after (verified both ways by reverting each file to HEAD in this worktree, confirming the new test fails, then restoring the fix and confirming it passes — never used git stash):
  • test/theme/tv_motion_profile_scroll_sites_test.dart — board card: snappy settles its scroll-follow within 5 pumpAndSettle frames (140ms shipped glide, unchanged); smooth needs strictly more (observed 6, driven by the real 260ms AppMotion.tvScroll figure).
  • test/detail_layout_rec_scroll_follow_motion_profile_test.dart — Marquee rec rail: snappy stays within 3 frames (unchanged instant jump); smooth needs at least 5 (observed 6) once it's actually gliding over 260ms instead of jumping. Console/Dossier share the identical fix and are exercised unmodified by the existing test/detail_layout_rec_scroll_follow_test.dart (still green — the scroll still lands, just now profile-aware).
• Ran the full requested suite plus the new tests together: test/theme/tv_motion_profile_test.dart, test/settings_tv_motion_row_test.dart, test/hover_grow_motion_test.dart, test/theme/focus_expression_motion_test.dart, test/theme/shape_type_motion_test.dart, test/detail_layout_rec_scroll_follow_test.dart, test/indexer_managers_settings_page_test.dart, test/pikpak_folder_picker_dialog_test.dart — all pass (105/105).
• Also ran the broader suites touching the changed files for extra confidence: test/detail_layouts_dpad_test.dart, test/detail_showcase_compact_test.dart, test/detail_showcase_test.dart, test/detail_theme_layout_test.dart (1071/1071), test/search_board_cell_origin_test.dart, test/search_public_types_pin_test.dart, test/theme/shape_manifest_test.dart.
• dart analyze clean on every changed and new file.

Risks

• Low. Every change is additive routing through an already-tested API (AppMotion.tvScroll / AppMotion.scrollTempo); the snappy-profile branch is written to reproduce the exact prior literal at each site, so default (snappy) behavior is unchanged.
• The two new tests infer animation duration from pumpAndSettle frame counts rather than sampling a fixed elapsed time, since the post-frame callback's start latency is not deterministic in flutter_test; thresholds carry a documented margin against that noise.
• PlatformUtil.isTelevision is a global flag read directly in three of the fixed sites (detail_layout_console/dossier/marquee.dart), consistent with the existing pattern in that file family — this can't be overridden per-widget in tests, so the new Marquee test uses PlatformUtil.debugSetAndroidTvCached(true), the existing test-only hook for this.
• Did not touch lib/utils/tv_reveal.dart (tvRevealMinimal, used by Settings pages like Indexer Managers/IPTV/Quick Play) or source_row.dart/torrent_result_row.dart/provider_status_cards.dart: none of these branch on isTelevision for their duration (same figure on every platform), so they aren't a differential "TV motion profile" regression the way the six fixed sites were — flagging them here rather than silently expanding scope.