fix: NetCDF forcing month-range off-by-one + CSV quantization matching

[DankerMu]

Summary

Fixes the off-by-one bug in NetcdfForcingProvider::initCmfdMonths() that caused a fatal error when END falls exactly on a month boundary (e.g. END=730 days from 20170101 → requests non-existent 201901 file).

Adds CSV-matching quantization in loadTimeStep() so that NetCDF forcing values are bit-identical to AutoSHUD-generated baseline CSV forcing.

Changes

NetcdfForcingProvider.cpp

• initCmfdMonths(): use nextafter(sim_end_min, -inf) to treat END as exclusive bound for file discovery
• loadTimeStep(): add rounding to match AutoSHUD CSV quantization (precip 4dp, temp 2dp, RH 4dp, wind 2dp + 0.05 min clamp, radiation integer)
• maxTimeMin(): account for step-function semantics — last record covers one more interval beyond its timestamp

TimeSeriesData.cpp/hpp

• Add getMaxTimeCovered() that returns maxTime + lastDt for step-function coverage

MD_readin.cpp

• Use getMaxTimeCovered() in validateTimeStamps() for CSV forcing coverage check

Verification

10-day QHH run (END=730, START=1):

• compare_forcing.py: all 5 variables max_abs_diff = 0 (exact match)
• Output .dat files: binary identical (md5sum match)
• Baseline rerun with NETCDF=1 binary: output unchanged

[DankerMu]

Review: SHUD PR #78

initCmfdMonths nextafter fix (L781-791)

Correct. END is exclusive; nextafter(sim_end_min, -inf) prevents requesting the month of the boundary day. Guard handles degenerate case.

loadTimeStep quantization (L1260-1325)

Rounding matches AutoSHUD R CSV output. Order correct: clamp NaN/negative → round → threshold. nearbyint is the right choice.

maxTimeMin step-function (L1762-1779)

Backward scan for last positive dt is robust. Correctly extends coverage for step-function semantics.

TimeSeriesData::getMaxTimeCovered

Clean addition. lastDtMin computed during file scan, no extra I/O.

Verification

10-day QHH: all 5 forcing vars max_abs_diff=0, output .dat binary identical, baseline unbroken.