refactor(linter): make rule execution declarative

[melonamin]

What changed

• pass dialect, token, and templating metadata through immutable rule contexts
• register each lint rule with declarative engine, scope/source, dialect, quality, and fallback metadata
• schedule and filter rules from descriptors instead of rule-code switches
• preserve the public lint configuration, LintContext, LintRule, and diagnostic contracts
• add architecture tests for descriptor completeness, filtering, context isolation, compatibility, and fallback behavior

Why

Rule execution metadata and scheduling policy were split between thread-local state and centralized rule-code switches. This made nested execution implicit and required scheduler edits whenever a rule changed.

Validation

• cargo fmt --all -- --check
• focused linter architecture tests
• cargo test -p flowscope-core
• cargo clippy -p flowscope-core --lib -- -D warnings
• codex review --uncommitted