Fixed longwave offset is what stops an energy-balance alignment being exact

[dfalster]

Not a new issue — this is #28 / PLAN item 13.2, with a concrete cost attached now
that 13.1 has landed in #93.

Aligning phylloptim's energy balance to Sicangco et al. (2026) for
leaf_calibration_test/sicangco-2026 works to 0.07 K in leaf temperature over
Tair 20–60 °C, which is better than expected. It gets there by setting Rn_ and
ra_ from their formulation after every set_drivers() call and running four
Picard passes, because two of their inputs depend on Tleaf (the Grashof term in
Gbhfree, and their isothermal net radiation through Dleaf).

What cannot be aligned is the longwave term. Theirs carries an atmospheric
emissivity that depends on air temperature and vapour pressure:

ema     = 0.642 * (ea / Tair_k)^(1/7)
Rnetiso = LeafAbs * Rsol - (1 - ema) * sigma * Tair_k^4

Ours is longwave_net_offset = -40 W m^-2, fixed. Over that sweep their term runs
roughly −98 to −133 W m⁻², so the fixed offset is out by 60–90 W m⁻² at the hot
end — which is why test_energy_balance_stomatal_decoupling has to override Rn_
by hand and says so at length.

Two notes for whoever picks #28 up:

• ⚠️ It makes Rn depend on Tleaf, so the explicit balance stops being
  explicit. Either plantecophys's radiation-conductance linearisation or an inner
  iteration. PLAN 13.2 already says this.
• The alignment above is the concrete test case: if Temperature-dependent outgoing longwave in the Penman-Monteith Rn (from plant #581 / #567 review) #28 lands, this example's
  Picard loop should collapse to one pass and the 0.07 K should improve. Worth
  re-running it as a check rather than inventing a new one.