Release 1.1.0: undefined-predicate check and docs overhaul

[YoanSallami]

Two commits: the CI smoke-test fix already on this branch, and the 1.1.0 release work.

Version bumped to 1.1.0 in Cargo.toml, Cargo.lock and pyproject.toml.

Verifier

• New undefined-predicate check (src/verifier/undefined.rs). A body reference to an undefined Predicate-cased name is silently treated as an external database table by the compiler, so Custmer(id:) does not fail — it reads a phantom table and returns nothing. The check flags it and, when a defined predicate is close enough by edit distance, suggests it.
• Lowercase references (raw tables) and function calls (legitimate SQL passthrough) are deliberately left alone.

Compiler and parser

• Collapse the unused Parse/Compile error variants into Syntax and Generic.
• Remove the incantation mode from the parser.
• Restore upstream Logica attribution headers on modified files; add NOTICE.

Docs

• New Why Synalog page: the business case, what changes in practice, what Synalog is not, and how it compares to text-to-SQL, BI semantic layers, graph databases, RDF stores and vector RAG.
• Knowledge graphs expanded with the modeling rationale: why model a graph at all, what it buys and what it costs, a comparison with a dedicated graph database, a "choosing a time model" decision guide, where bitemporality pays for itself and what it costs.
• New runnable examples: temporal_graph.l and bitemporal.l.
• Regenerate example outputs; remove CLAUDE.md.

Verification

• cargo test: all suites pass (735 + 14 + 18 + 1 + 1, 0 failed, 2 ignored).
• zensical build: no issues.

Note: the branch name predates this work — it was originally just the CI smoke-test fix.

🤖 Generated with Claude Code