fix(hapi): replace assert-based input validation with raises

[MAfarrag]

Context

src/hapi used assert for input validation in 53 places. assert is stripped under
python -O, so none of those were doing real validation -- an invalid input sailed past the
guard and failed further in, on a shape mismatch or a None. PR #214 converted every one. This
issue documents that change and the defects it surfaced along the way.

What Changed

• Every assert in src/hapi doing input checking now raises TypeError (wrong type) or
  ValueError (wrong value/length) instead. src/hapi now contains no assert.
• Repeated checks were pulled into helpers instead of becoming dozens of near-identical if
  blocks: _check_parameters_cover_grid and _check_lake_meteo in run.py,
  _check_optimization_args in calibration.py.
• Three messages that were already wrong, independent of the -O issue, are corrected rather
  than carried over verbatim (the store_history/history_fname messages in
  run_calibration/FW1Calibration, lumpedCalibration's basic_inputs message, and the
  four maxbas checks' wording).

Defects found while doing this

• Catchment.__init__ now canonicalises and rejects routing_method. It used to store
  the string verbatim, so a lower-case "muskingum" silently routed every cell down the
  MAXBAS branch (distrrm.SpatialRouting compares != "Muskingum" case-sensitively) and
  raised TypeError on bankfull_depth, which is None outside the flood model. Breaking
  change: landed as refactor(catchment)! with a BREAKING CHANGE: footer.
• read_discharge_gauges filled QGauges by the wrong key -- labelled from
  gauges.column but filled by int(id), so any column != "id" produced the requested
  columns entirely NaN plus a second, id-named set beside them, silently.
• Routing.calculate_weights accepted a MAXBAS below one -- 0.5 produced an all-zero
  hydrograph with nothing raised. Fixing it exposed a real defect in the lake test fixtures,
  which had been routing on the wrong parameter set and asserting against empty output.
• save_results concatenated its output directory instead of joining it, so some/dir
  without a trailing separator wrote some/dirResult_2009-01-01.tif beside the directory
  rather than inside it.
• Three methods advertised str | dt.datetime but called strptime unconditionally
  (plot_hydrograph, read_discharge_gauges's split=True path, save_results); each now
  branches on isinstance(..., str).

Affected locations

File	Notes
src/hapi/catchment.py	routing_method canonicalisation, QGauges fix, save_results fix, datetime branches
src/hapi/routing.py	calculate_weights guard
src/hapi/calibration.py	_check_optimization_args, two Sonar-flagged guards
src/hapi/run.py	_check_parameters_cover_grid, _check_lake_meteo
src/hapi/inputs.py, src/hapi/rrm/*.py	remaining assert -> raise conversions

Verification

• Every converted check verified to raise under both python and python -O
• Reproduced the gauges.column corruption, the all-zero MAXBAS hydrograph and both
  strptime bugs, then confirmed each fixed
• Full suite green (554 tests) after the conversion
• Two full /review-rounds passes and a SonarCloud sweep (23 open issues down to 6)

Implemented in #214.