Fix stale build, parameter, and CI claims in docs and comments

[sbryngelson]

• Generated-file header and NAMELIST_VARS comment said to regenerate via a
  cmake reconfigure; generation is a build-time custom command
  (cmake/ParamsCodegen.cmake).
• copilot-instructions told reviewers to flag PRs that do not hand-edit the
  namelist/m_global_parameters, which are now generated from definitions.py;
  point at NAMELIST_VARS instead. Same fix in architecture.md.in.
• contributing.md: drop the stale "15 files" count; '#:if MFC_SIMULATION'
  is a Fypp NameError in pre/post_process and false in simulation (fypp -D
  without a value binds None), so use '#:if defined('MFC_SIMULATION')';
  cleanliness.yml only reports warning increases and never fails, so say so
  and drop the "Warnings increased" CI-failure row; the PR template has no
  GPU checklist, so point to gpuParallelization instead.
• getting-started.md: 'params --family' is rejected by argparse; the flag
  is --families.
• gpuParallelization.md: '$:$GPU_PARALLEL_LOOP' typo.
• Parameter counts (~3,300 / ~3,400 / over 3,000): './mfc.sh params' reports
  4211. Docs now say "over 4,000"; code docstrings and CLI help drop the
  number.

Verified: fypp from build/venv on a scratch file reproduces the NameError and
the None-is-false behavior of '#:if MFC_SIMULATION', while the defined() form
selects the right branch per target; './mfc.sh params --families' works and
'--family' errors; params_tests and lint_docs pass.