Fix event-loop starvation, MeshCore stability, config-page hardening

- mesh_data_store.py / env/store.py: make refresh() async, offload blocking
  polls via asyncio.to_thread/gather so 7 lockstep sources no longer starve
  the shared event loop.
- main.py: gather pollers concurrently + set_default_executor thread pool.
- Dockerfile / docker-compose.yml: healthcheck now curls the dashboard for a
  real liveness signal instead of a process-exists check.
- transport/meshcore_transport.py: MeshCore keepalive loop (get_time() every
  120s), reconnect re-arm (_post_reconnect_setup_async from
  _on_connect_event), and MC channel-name normalization
  (_resolve_mc_channel_idx strips a leading #).
- dashboard-frontend: MeshCoreConnection.tsx config-page hardening, new
  ErrorBoundary component, wired into App.tsx.
- tests: fix ~40 call sites broken by refresh() becoming async (
  test_generic_http.py, test_store_received_delta.py,
  test_store_wzdx_persist.py) by wrapping with asyncio.run(), matching this
  suite's existing convention for calling async code from sync test
  functions. Verified: all 40 tests pass.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

work/tests/test_generic_http.py

@@ -8,6 +8,7 @@ Ported-behavior coverage:
   * geometry-path Point -> centroid extraction
 """
 from __future__ import annotations
+import asyncio
 
 import json
 
@@ -203,7 +204,7 @@ def test_cold_start_silent_first_poll_seeds_persists_no_emit():
     # Stub the network fetch with one active outage.
     adapter._fetch = lambda url, extra_headers=None: _payload([IDAHO_POWER_ITEM])
 
-    store.refresh()  # poll 1 == pre-existing backlog
+    asyncio.run(store.refresh())  # poll 1 == pre-existing backlog
 
     # Nothing broadcast on the cold-start poll...
     assert captured == [], "first poll must broadcast NOTHING (cold-start seed)"
@@ -220,14 +221,14 @@ def test_cold_start_silent_first_poll_seeds_persists_no_emit():
 def test_later_poll_broadcasts_newly_received_item():
     store, adapter, captured = _make_store_with_generic()
     adapter._fetch = lambda url, extra_headers=None: _payload([IDAHO_POWER_ITEM])
-    store.refresh()  # poll 1 — seed silently
+    asyncio.run(store.refresh())  # poll 1 — seed silently
     assert captured == []
 
     # A genuinely NEW outage appears on a later poll -> it must broadcast.
     new_item = dict(IDAHO_POWER_ITEM, omsOutageId="456", omsCustomerCount=99)
     adapter._fetch = lambda url, extra_headers=None: _payload([IDAHO_POWER_ITEM, new_item])
     adapter._last_poll.clear()  # force cadence to elapse
-    store.refresh()  # poll 2
+    asyncio.run(store.refresh())  # poll 2
 
     assert len(captured) == 1, "only the newly-received outage broadcasts"
     assert captured[0].category == "power_outage"
@@ -367,7 +368,7 @@ def test_build_generic_detail_reader():
     from meshai.notifications.env_reporter import EnvReporter
     store, adapter, captured = _make_store_with_generic()
     adapter._fetch = lambda url, extra_headers=None: _payload([IDAHO_POWER_ITEM])
-    store.refresh()
+    asyncio.run(store.refresh())
 
     text = EnvReporter().build_generic_detail()
     assert "idaho_power" in text