Fix stale config names (c17/l17) that break the README quickstart

[artlbv]

Problem

Following the README quickstart on a fresh clone, every law run command fails with

ValueError: object 'l17' not known to index 'UniqueObjectIndex(cls=order.config.Config, len=10)'

The configs were renamed to c17v9 / c22prev14 / ... in fdab666, but several default_config
values and two hardcoded training_configs still refer to the old c17 / l17 names. Since none
of the quickstart commands pass --config, they all fall back to those defaults.

law.dl.nocert.cfg is the one that bites first, because it is the default answer for
HBW_LAW_CONFIG in setup.sh.

Changes

Faithful rename of the leftovers, c17 → c17v9 and l17 → l17v9:

• law.cfg, law.shared.cfg, law.sl.cfg, law.sl.nocert.cfg
• law.dl.nocert.cfg, law.resonant.nocert.cfg
• examples/000_test/common.sh
• notebooks/setup.py
• hbw/ml/old_ml_model.py, hbw/ml/derived/legacy_dl.py

And, so the quickstart actually runs end to end:

• pass an explicit --config in the three law run examples
• document that hbw.BuildCampaignSummary has to be run once per config first — the lazy config
  factory in hbw/analysis/create_analysis.py instantiates that task and refuses to build the
  config until its output exists, so without it every task fails with
  "Campaign used for ... is not yet initialized"
• drop --producers features from the plotting example (hbw.production.features is not in
  production_modules any more)
• replace the config_2017 / config_2017_limited paragraph with the list of configs that exist,
  and fix the dead link to hbw/config/analysis_hbw.py

Testing

tests/run_linting passes. Verified separately that with these names the analysis resolves its
configs instead of raising, and that hbw.BuildCampaignSummary is indeed required before anything
else.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MdTWv3sC68nLDdYw12GV7K