[codex] Change battery actions to policy states

[LordMike]

Issue

WattPlan exposed battery action states that looked like raw flow states (hold, discharge, charge_pv, charge_grid_pv). That made hold especially misleading: the planner could mean "no modeled battery flow", while inverter automations often interpret it as "block discharge". In a real system, PV may cover forecast load while the battery should still be allowed to cover unexpected load spikes.

Fix

This changes the public battery action contract to inverter-control policy states:

• preserve: prevent battery discharge because the model shows spending energy now is bad or infeasible.
• self_consume: normal operation; allow battery discharge for real load and do not request grid charging.
• grid_charge: request/allow grid charging and avoid spending the battery while charging.

PV charging remains implicit/normal behavior. PV export is intentionally not represented as a battery action state and is deferred for future site-level policy design.

What Changed

• Removed old public action states from sensor options, tests, docs, and plan projection mappings.
• Updated HA projection action codes to gc, p, and sc for the three policy states.
• Removed user-facing prefer_pv_surplus_charging configuration/docs where it conflicted with the policy model, leaving it documented only as internal/deferred optimizer behavior.
• Added a model-backed battery_preserve signal to optimizer state and warm-start reuse.
• Replaced target/future-flow preserve inference with a counterfactual MPC check: if forcing a small discharge now is infeasible or worsens the objective, the slot is marked preserve; otherwise it defaults to self_consume unless grid charging is active.
• Updated docs with the new policy contract and a generic Solar Assistant/MQTT/Deye-compatible automation example.

Validation

• ./scripts/run_tests.sh
• Result: 135 passed in 4.65s

Caveats

• preserve now requires additional counterfactual solver calls for candidate battery slots.
• Targets do not directly create preserve; they only affect preserve through model constraints/objective impact.
• PV export remains deferred as a site-level policy, not a battery state.