Add structured semantic evaluation feedback

[rlippmann]

Goal

Expose structured semantic evaluation information so downstream consumers can
act on compiler results without parsing human-readable error messages.

Motivation

The engine already determines why a canonical directive cannot be applied.
Those reasons are currently rendered primarily as error message strings.

Consumers such as Directive Drafter need machine-readable evaluation results
to repair candidate directives. Parsing error messages would couple consumers
to presentation text rather than compiler semantics.

Scope

Preserve existing semantic behavior while making evaluation results available
in structured form.

Examples of existing outcomes that may need structured representation:

• conflicting policy state;
• missing replacement preconditions;
• invalid state transitions;
• already-satisfied transitions where additional context is useful.

Relationship to #220

The transition evaluation boundary introduced by #220 provides the natural
location for structured evaluation feedback.

This issue does not define new transition semantics. It exposes existing
evaluation outcomes in a form that downstream consumers can use.