refactor: canonicalize directive text boundary

[rlippmann]

What changed

• canonicalized CanonicalDirective.text as a computed property derived from kind and operands
• updated grammar docs, API contract fixtures, and grammar tests so directive text is treated as canonical serialized syntax rather than preserved caller input
• fixed normalization and replacement property generators so they only assert on true normalized-key equivalence and canonical replacement inputs
• added focused grammar coverage for the unsupported canonical-serialization kind branch to bring grammar coverage to 100%

Why

• CanonicalDirective is a public grammar object, so its surface should reflect semantic canonical serialization rather than historical parser provenance
• property-based engine tests should use the actual normalization and grammar rules as their oracle instead of relying on looser assumptions
• the grammar module is intended to be a tightly specified boundary, and fuller coverage helps keep that contract stable in CI

Checklist

• pre-commit run (uv run pre-commit run --all-files)
• tests pass (uv run pytest)