simulate: evaluate CEL guards from a supplied --context

[joshua-temple]

Follow-up to #175 (crucible simulate).

#175 ships per-guard verdict seeding (-guard name=true/false, default false) so guarded transitions can be driven headlessly without Go bindings. That covers Go-func guards, but it's a manual verdict, not real evaluation.

For IRs whose guards carry CEL expressions (state/expr), simulate could instead evaluate guards live against a user-supplied context:

crucible simulate machine.json -events "checkout,paid" -context '{"items": 3, "paid": true}'

This would wire a celGuard per CEL-authored guard, project the supplied JSON context through the machine's ContextSchema activation, and resolve guarded transitions from real data instead of seeded verdicts.

Scope notes:

• Only helps guards authored as CEL expressions; Go-func guards still need -guard seeding.
• Requires schema detection on the loaded IR and the state/expr dependency in cmd/crucible.
• -guard overrides and -context should compose (explicit -guard wins).