test(provider): make the syntaxError classification exhaustive (#591)

[e54-bot]

Summary

• Adds EXCLUDED_CODES (every emitted diagnostic code that must not produce syntaxError, grouped with documented deferral reasons) and NON_DIAGNOSTIC_LITERALS alongside DOCUMENT_SYNTAX_CODES.
• Adds syntax_error_classification_is_exhaustive: scans all kebab-case string literals in crates/opy-rs/src (excluding test dirs), skipping comments/raw strings, and asserts every emitted code-shaped literal appears in exactly one of the three lists — unclassified new codes and stale listed codes both fail.
• Directly implements Escalation: scope-defining allowlist merged incomplete, omissions silently accepted #591's prevention proposal: the allowlist went stale twice (fix(provider): count string-entity lexer errors as validateEdits syntaxError #587 string entities, fix(provider): count directive-malformation and enum-increment as syntaxError #588 script/enum codes); classification is now executable rather than review-memory.

Test plan

• cargo test -p opy-provider — 69/69 incl. the new exhaustive test
• cargo fmt --all -- --check, cargo clippy -p opy-provider --all-targets -- -D warnings, git diff --check

[Teakowa]

Exhaustive classification verified against merged main post-#592: all emitted kebab-case literals classify exactly once, no stale entries, test fails on unclassified additions. CI 6/6 green.