[BUG] Heat-pump balance slacks (`phss`, `phps2`) are free in opf_version 1/3 — heat demand can be served by costless phantom power

[joda9]

Details

• variable_slack_heat_pump_storage (→ phss, phps2, both ≥ 0, no upper bound) is
  created for all opf_versions (src/prob/opf_bf.jl).
• The heat-pump balance holds in all versions (src/core/constraint.jl):
  cop·(php + phps2) == pd + phss − phs.
• The penalty term 1e4·Σ(phss + phps2) exists only in the objectives of version 2 and 4
  (objective_min_losses_slacks, ..._OG). The version 1/3 objectives
  (objective_min_line_loading_max, ..._OG) contain no term for them.

Consequence

In opf_version 1/3 (1 is the default), phps2/phss are free degrees of freedom: the optimizer
sets php = 0 (heat pump off — less load means fewer losses and lower loading, so this is
optimal) and "serves" the entire heat demand via phps2 — phantom electrical power that never
appears in any power balance. The returned "optimized" heat-pump schedule is thermally
meaningless while looking electrically plausible.

Intended semantics / fix

These slacks are a data-consistency valve (heat demand exceeding heat pump + storage
capability), not a grid-restriction slack — the risk they guard against is independent of the
opf_version. Fix accordingly:

1. Penalize phss/phps2 very expensively in all four objectives (big-M style, like the
   existing 1e4 in v2/4), so they are used only when physically unavoidable.
2. Emit a warning whenever the valve is used beyond numerical tolerance (heat pump, time
   step, missing energy) — silent absorption is explicitly unwanted.