feat(planner): add EV SoC economics sensor comparing cost across target SoC and deadlines

[woopstar]

Summary

Adds an "EV SoC Economics" diagnostic sensor per EV that answers "what would it
cost to charge to X% by deadline Y?" — the real-money cost of charging to each
of 50/60/70/80/100% SoC by the next occurrence of 08:00 and by the next
occurrence of 17:00.

Each combination is computed by cloning the coordinator's last-used
PlannerInput, overriding only that EV's target-SoC/deadline fields
(ev_planned_load_target_soc_pct / ev_planned_load_deadline, or the
ev_second_* equivalents), and re-running the existing pure
run_planner(PlannerInput) -> PlannerOutput. PlannerOutput.plan_cost.total_cost
is read back per combination. No planner/MILP semantics change — this
calls the existing pure engine unmodified with different inputs.

No auto-recommended target is exposed — the raw cost/delta numbers are
surfaced and the user decides.

What changed

Planner (pure, no HA imports):

• custom_components/hsem/planner/ev_soc_economics.py — next_time_of_day()
  helper, EVSoCEconomicsPoint/EVSoCEconomicsResult dataclasses, and
  compute_ev_soc_economics(). Guard clauses mirror
  ev_planner.build_ev_charging_plan()'s early-outs (not
  enabled/connected/smart-charging-disabled/zero capacity). Targets at or
  below the EV's current SoC cost 0.0 and skip solving entirely. Feasibility
  is a simple charger-power-based check independent of price.

Coordinator:

• custom_components/hsem/coordinator_ev_soc_economics.py — new mixin, called
  from coordinator_cycle.py right after _run_planner_phase() returns.
  Throttled to at most every 30 minutes
  (EV_SOC_ECONOMICS_RECOMPUTE_MIN_SECONDS), requires
  self._last_planner_input to be set, solves both EVs in a single
  hass.async_add_executor_job() call, and re-checks the update-generation
  guard afterwards (same stale-cycle pattern as the main run_planner() call).
• New fields on CoordinatorData / CoordinatorSharedState:
  ev_soc_economics, ev_second_soc_economics.

Sensors:

• custom_sensors/ev_soc_economics_sensor.py (HSEMEVSoCEconomicsSensor) and
  custom_sensors/ev_second_soc_economics_sensor.py
  (HSEMEVSecondSoCEconomicsSensor) — mirror the existing
  ev_optimal_charging_plan sensors exactly (DIAGNOSTIC, RestoreEntity,
  ENUM device class). State is one of ready / not_connected /
  smart_charging_disabled / unavailable; the full cost/delta table lives
  in extra_state_attributes via EVSoCEconomicsResult.as_attributes().
• New naming helpers in utils/sensornames/ev.py, registered in sensor.py
  under the existing hsem_ev_planned_load_enabled /
  hsem_ev_second_planned_load_enabled gates.
• Translation keys added to translations/en.json and translations/da.json.

Dashboard:

• Conditional markdown-table card per EV in
  custom_components/hsem/dashboards/dashboard_en.yaml's "EV Charging"
  section, rendering the points attribute grouped by deadline_label via
  Jinja's groupby filter (not custom:apexcharts-card — this is a small
  categorical table, not a time series). Mirrored into
  docs/dashboard-setup.md's "Full dashboard YAML" block; card-count table
  and "EV Charging tiles" reference table updated.

Docs:

• New "EV SoC economics sensors" subsection in docs/sensors-reference.md
  under "EV charging plan sensors". docs/planner-spec.md is untouched — no
  planner semantics changed.

Tests

• tests/planner/test_ev_soc_economics.py — next_time_of_day(), guard
  states (no run_planner() call), monotonic cost per deadline column,
  zero-cost/no-solve for already-met targets, price-independent feasibility,
  delta fields, second-EV field isolation (17 tests).
• tests/test_coordinator.py — throttle window, first-call compute,
  no-op without _last_planner_input, single executor job for both EVs,
  stale-generation discard (6 tests).
• tests/sensors/test_ev_soc_economics_sensor.py — state/attributes/
  availability/restore for both sensors (25 tests).
• Fixed two pre-existing tests/test_ha_mock_integration.py bare-coordinator
  helpers/mocks that didn't know about the new coordinator fields/executor
  call.

Test and lint results

All four quality gates pass:

./scripts/quality.sh lint     # ruff format + ruff check + prettier — clean
./scripts/quality.sh typing   # mypy — 0 errors
./scripts/quality.sh quality  # pyright + vulture — 0 errors
./scripts/quality.sh test     # 3088 passed

Acceptance criteria (from #903)

• compute_ev_soc_economics() returns a monotonically non-decreasing cost
  per target within each deadline column, for a fixed price/PV forecast.
• Targets at or below the EV's current SoC cost 0.0 and do not trigger
  an extra run_planner() solve.
• feasible is False when the charger's max power cannot physically
  reach the target by the deadline, independent of price.
• Guard states (not_connected, smart_charging_disabled,
  unavailable) short-circuit without calling run_planner() at all.
• Recomputation is throttled and stale-update-cycle safe.
• The second-EV sensor only appears when
  hsem_ev_second_planned_load_enabled is set.
• docs/sensors-reference.md documents both new sensors.
• Bundled dashboard shows a conditional markdown-table card per EV,
  mirrored into docs/dashboard-setup.md.
• Tests added covering the above.
• ./scripts/quality.sh all passes.

Fixes #903