Add Roku TV as a digital frame driver (cast via HA core roku media_player)

[dsackr]

Roku has no local API to receive arbitrary image bytes, so this driver
targets an existing HA core roku media_player entity and casts images
via media_player.play_media, staging the PNG behind a short-lived
HA-hosted token URL (same pattern as the Samsung MDC driver) since HA's
Roku integration launches its built-in Media Player app off a URL rather
than accepting bytes directly.

Adds config_flow add_roku step (entity picker, no discovery/probe),
RokuCoordinator, wiring through init/http_api/panel_codec/helpers/sensor/
library_http, panel + strings updates, KPF 39 (renumbered from 36 after
merging main, which had already taken 36-38), and backend tests.

Co-Authored-By: Claude Sonnet 5
Claude-Session: https://claude.ai/code/session_01KmCpFs3Y4KhAjap7dSMNco

[dsackr]

Pushed a fix for the validate (hassfest) failure — after_dependencies wasn't alphabetically sorted with the rest of manifest.json's keys.

The test job's 3 pytest errors (not failures — 544 passed, 3 errors) are a pre-existing aiodhcpwatcher/scapy socket issue, not something this PR introduces: I confirmed it reproduces identically (527 passed, 3 errors) on a clean checkout of main, and docs/TEST_LEDGER.md already shows main red with this same "N passed, 3 errors" signature going back several commits (e.g. 185c654). Leaving it alone here since it's unrelated to the Roku driver; will re-run once main's CI recovers.

---

Generated by Claude Code

[dsackr]

The panel-playwright failure (2 failed, 166 passed) is also pre-existing on main, not something this PR introduces — the 2 failing specs are addons-catalog-refresh.spec.js and addons-categories.spec.js (scene-pack/add-ons catalog UI), unrelated to the Roku driver or anything this PR touches. docs/TEST_LEDGER.md shows the identical 2 failed signature on every panel-playwright CI run on main going back to at least 60f6213 (2026-07-27).

So both currently-red checks on this PR (test/backend-pytest's 3 errors, test/panel-playwright's 2 failures) are carried over from main being red, not regressions from this change. I'll rebase onto main and re-check once those recover.

---

Generated by Claude Code

[dsackr]

Update: main moved substantially since the comments above (v0.12.171 → v0.12.179), fixing panel-playwright and most of the backend flakiness along the way. This branch had drifted into a real merge conflict against main (in __init__.py, between this driver's dispatch and a new 31.5" resolution migration) — resolved and pushed as a merge commit (a54dcbc).

Current state: validate ✅, panel-playwright ✅, backend-pytest still red but down to exactly one pre-existing test — tests/python/setup/test_init_setup_entry.py::test_late_bound_tracker_watch_does_not_violate_thread_safety (pytest_socket.SocketBlockedError via aiodhcpwatcher/scapy) — same root cause flagged in my first comment, now isolated to a single test. Confirmed this reproduces identically on a clean main checkout, so it's not something this PR introduces. Will merge main again once that test is fixed upstream.

---

Generated by Claude Code

[dsackr]

Pushed the same DHCP/scapy socket fix here too (a7d67c4) — CI now confirms 2 failed, 565 passed, 0 errors. The DHCP test error that's been blocking this PR for days is resolved; the only remaining failures are the pre-existing, unrelated "Daily Newspaper" skill-count regression on main (test_built_in_skills_seeded_on_first_load, test_built_ins_not_reseeded_after_user_deletes_one) — not something this PR touches. Also opened #21 to land the DHCP fix on main itself.

This PR is functionally green modulo that one unrelated pre-existing issue.

---

Generated by Claude Code

[dsackr]

CI confirms: all checks green (validate ×2, backend-pytest, panel-playwright) on the merge commit 2ee5e1f. This PR is now fully mergeable and ready for review — no known issues remaining.

---

Generated by Claude Code