Fix excessively negative ET over sandy soil in eCLM-ParFlow coupling (#105)

[s-poll]

Fixes #105. In coupled eCLM-ParFlow runs, sandy soil columns in very dry conditions produced unrealistically large negative ET (100–1000 W/m² day and night). Root cause: when COUP_OAS_PFL is active, the ParFlow-derived pressure head (pfl_psi) is used directly as the soil matric/suction potential without the lower-bound clamp that the standalone (non-coupled) code path already applies via smpmin. In dry sand, ParFlow's pressure head can reach extremely large negative values, which fed straight into eCLM's humidity and water-movement calculations as an unbounded suction potential — effectively telling eCLM the soil was far drier than the model's own retention-curve physics would allow, driving the excessive ET.

Changes
src/clm5/biogeophys/CanopyTemperatureMod.F90: psit (used for soil surface relative humidity) is now clamped the same way the uncoupled branch already does: psit = max(smpmin(c), min(0._r8, pfl_psi(c,1))), instead of passing pfl_psi(c,1) through unbounded.
src/clm5/biogeophys/SoilWaterMovementMod.F90: smp_l(c,j) is likewise clamped to smpmin(c) when set from pfl_psi(c,j), instead of being assigned the raw coupled pressure head.
src/clm5/biogeophys/SoilStateInitTimeConstMod.F90: smpmin_col is initialized to -1.e6_r8 mm under COUP_OAS_PFL (vs. the standalone default of -1.e8_r8 mm), giving the coupled pathway a physically tighter floor appropriate to the coupling exchange rather than the very permissive uncoupled limit.
Together these ensure ParFlow-communicated pressure heads are bounded the same way eCLM already bounds its own internally-derived suction potentials, closing the gap that let the coupling interface effectively communicate "unlimited water availability" in dry sandy conditions.