[DOCS] ADR for evaluation details

We need to add a retrospective ADR for the evaluation details based on the evaluation state.

It should be covered in an ADR, but can probably be reduced to a single comprehensive mermaid flow-chart, which covers how the evaluation details should be defined based on the evaluation flow. Particularly, the REASON should be captured:

• when a variant is defined, and no or empty targeting is specified, the REASON=STATIC
• when a targeting rule returned a variant, the REASON=TARGETING_MATCH
• when a targeting rule returns null, the REASON=DEFAULT and the defaultVariant is returned
• when there's an error in evaluation, REASON=ERROR and ERROR_CODE={some-error-code}
• when a value is returned from CACHE, REASON=CACHE see caching details here; keep in mind this is only relevant for RPC mode
• all error code behavior: PROVIDER_NOT_READY, FLAG_NOT_FOUND etc, and precisely what flows and states result in which errors

Start by tracing the evaluation flow, including variant definition, targeting results, default fallback, errors, and RPC-mode cache behavior described in the issue. Use the linked caching details and enumerate the stated error codes and resulting states. Done means an ADR with a single comprehensive Mermaid flowchart covering each evaluation-detail and error path.