Add soil property perturbation script for eCLM ensemble generation

[jjokella]

Solves #72

Provides perturbed surface files for
https://hpscterrsys.github.io/eCLM/users_guide/introduction/soil_hydraulic_parameters.html

Summary

• Add perturb_soil_properties.py for generating perturbed eCLM
  surface file ensembles in two modes: hydraulic (Clapp-Hornberger
  pedotransfer functions with organic matter adjustment, perturbing
  PSIS_SAT, THETAS, SHAPE_PARAM, KSAT) and texture (additive uniform
  noise on PCT_SAND, PCT_CLAY, ORGANIC)
• All paths configurable via CLI; output filenames derived from input file
  stem with zero-padded ensemble index
• Perturbation spreads (Gaussian std devs, uniform noise half-range, clipping
  bounds) and number of perturbed levels exposed as CLI arguments with
  physically motivated defaults
• --no-adj flag to write parameters without _adj suffix, selecting between
  soil_hyd_inparm_from_file_adj (default, all nlevgrnd layers, overwrites OM
  mixing) and soil_hyd_inparm_from_file (first nlevsoifl=10 layers, undergoes OM
  mixing)
• --n-perturb-levels to restrict perturbation to the uppermost N layers;
  deeper layers receive unperturbed CLM mean values
• Reproducibility via --seed (default: 42) and --state-file for
  saving/restoring RNG state to support resuming interrupted runs
• Original author credited: Yorck Ewerdwalbesloh
• Add README.md with full option reference and usage examples

Needs testing

• Run --mode hydraulic twice with the same --seed and verify byte-identical
  output
• Run --mode texture and verify sand + clay ≤ 100 % and no negative values
  throughout output
• Interrupt a run, resume with --state-file and --start N, verify no gap or
  overlap in member numbering
• Verify output NetCDF attributes contain correct perturbed_by (from $USER)
  and perturbed_on_date
• Run with --no-adj and verify variable names lack _adj suffix and are
  accepted by eCLM with soil_hyd_inparm_from_file = .true.
• Run with --n-perturb-levels N and verify levels ≥ N contain unperturbed CLM
  mean values