v0.13.0 — heatpumpmodel 0.5.0: the DPE mapping fix, and the two-branch A/W curve

Adopts heatpumpmodel 0.5.0, which carries two corrections pulling in opposite
directions: the #6 DPE mapping fix ("air" emission on a hydronic generator is a
fan-coil water loop at 45 degC, not the room at ~20 degC) and, because 0.5.0
sits on 0.3.0, the two-branch A/W COP curve (thesis Eq. 3.8 as printed, where
this project shipped only its defrost-degraded branch).

Measured per building on 849 heat-pump buildings over 19 national districts,
the same dynamic path run twice (v0.12.0's lock vs this release):

  A/A, any emitter      608   2.554 -> 2.554    0.000 %   control
  G/W + "air"            50   7.991 -> 5.037  -36.96 %   the mapping fix
  A/W, emitter != air   117   2.828 -> 3.128  +10.62 %   the curve
  A/W + "air"            74   4.551 -> 3.737  -17.88 %   both
  pooled                849   2.850 -> 2.828   -0.75 %   nothing

The pooled figure is a warning, not a result: it reads as 'nothing happened'
while 241 of 849 buildings moved by -37 % to +10.6 %. Both isolated figures
match upstream's own predictions (-40...-43 %, +4.4...+19.5 %).

Every published A/W number moves. default_heat_pump_config() no longer
reproduces Fig. 4's golden SCOPs; cop_curve=LEGACY_A_W_SINGLE_CURVE reproduces
the pre-0.13.0 A/W numbers exactly. cop_fl now takes (dT_hp, t_so) and raises
for A/W without it.