Query Workshop element facts and spans from workshop-rs instead of copying them #1006

Problem

wright-analyzer copies canonical Workshop facts and syntax that workshop-rs owns, which AGENTS.md forbids ("Never duplicate canonical Workshop data or semantics in Wright"):

• Element-name lists in crates/wright-analyzer/src/canonical/analysis.rs:
  • VOLATILE: random and time values.
  • ELEMENT_BINDER_VALUES.
  • The re-evaluated-argument names waitUntil, loopIf, ..., plus domain.ends_with("Reeval") and a "NONE" default parsed from a string.
  • canonical/cfg.rs: the wait 0.016 limit.
• A second Workshop lexer. SourceScan (analysis.rs) skips comments, matches parentheses backwards, and compares raw text against Else, If and End to compute fix spans for dead_branch_fix and evaluate_once_fix. It does not handle localized keywords.

When workshop-rs adds or renames an element, these lint rules can become wrong without any error.

Scope

Wright-side migration: query these properties and spans from workshop-rs, then delete the local lists and SourceScan.

Non-goals

Designing the workshop-rs API. That belongs to workshop-rs.

Dependencies

Blocked on workshop-rs:

• Expose element evaluation properties in the catalog workshop-rs#580: catalog element evaluation properties (nondeterministic values, per-element binder arguments, action-level re-evaluated arguments, typed enum defaults).
• Expose the full authored extent of nested condition and argument values workshop-rs#581: the full authored extent of nested condition and argument values, which replaces SourceScan.

The Wait minimum (0.016 in canonical/cfg.rs) is a measured runtime fact, not catalog data. Its source belongs to athena and is outside these two blockers.

Acceptance criteria

• wright-analyzer contains no hard-coded Workshop element-name lists and no text scanner for Workshop syntax.
• Existing lint and fix tests pass, and a localized-keyword fixture produces the same fix spans as the English one.