RRLS Studio v0.11

New: normalized support-gated predictive relations.
Run on an existing RRLS/Common Crawl SQLite corpus:

  py rrls_predictive_cli.py RRLS_CommonCrawl.sqlite --out predictive_report.json

The predictive analyzer is intentionally stricter than the earlier research notebooks:
- promotion is by whole context, never by the next symbol being predicted;
- every promoted context defines a normalized distribution;
- actual serialized model bytes are charged;
- host-disjoint held-out data is used by Common Crawl mode.

See RRLS_v11_RESEARCH_REPORT.md.
