#3875 Docs: document workflow evaluation counts and subphase timings

Swamp-club #3874 adds diagnostics to the successful swamp.workflow.evaluate span while preserving workflow.expressions_evaluated as the number of distinct successfully evaluated raw expressions.

Update content/manual/reference/opentelemetry.md, which currently lists only workflow.name and workflow.expressions_evaluated for that span:

• workflow.expression_occurrences: authored occurrences after the provenance gate, before runtime deferral.
• workflow.expressions_distinct: distinct raw expressions in those occurrences, including deferred ones.
• workflow.cel_evaluations: actual CEL attempts, including handled foreign-template failures.
• workflow.model_keys: model namespace keys, including both name and UUID aliases.
• workflow.cel_duration_ms: cumulative time in the CEL evaluation phase, including handled failures.
• workflow.provenance_duration_ms: cumulative spliceOrigin time, excluding subsequent substitution.

Repeated occurrences still evaluate independently. Subphase timings need not sum to the whole span. Fatal errors retain the error span without success diagnostics. No raw expressions, values or model names are exported by these new attributes. The in-repo design/enablers/expressions.md contains the same definitions.