Split structured runtime logging, trace capture, and dependency-graph diagnostics out of #41

[cwinland]

Docs reference: docs/roadmap/generator-roadmap.md

The current #41 branch work produced a reusable diagnostics snapshot and export surface. The remaining structured runtime logging and tracing work is larger and should be tracked separately so #41 can stay scoped to the shippable v4-2 diagnostics slice.

Scope

• define the structured runtime event model for setup, execution, and verification diagnostics
• define dependency-graph output beyond the current snapshot export surface
• define reusable call-trace capture for scenario and verification flows
• define sink, verbosity, and filtering behavior for the broader observability stack
• keep this design reusable for later helper normalization instead of creating per-feature diagnostics silos

Acceptance checklist

• The remaining runtime logging and tracing work is documented separately from the reusable snapshot/export work left in #41.
• Structured setup, execution, and verification event capture is explicit enough for implementation planning.
• Dependency-graph and call-trace outputs are defined as distinct deliverables instead of implied follow-on work.
• Sink, verbosity, and filtering requirements are explicit enough that later helper families can reuse them.
• The boundary with #43 remains explicit: this issue owns observability architecture, not container lifecycle safety.

Follow-on boundaries

• #41 stays focused on the reusable diagnostics snapshot/export slice for v4-2.
• broader helper normalization in #134 can depend on this issue instead of re-describing observability requirements.