Nest child group spans under their dependent group's span (#172)

[rubberduck203]

Summary

• RunGroups::execute parented every group's tracing span directly to doctor run regardless of requires, so dependencies showed up as siblings after their dependent instead of nested inside its span (Telemetry: Child group spans aren't properly nested #172). Replaced the flat traversal with a RunGraph — a DAG reconstructed from the planned run order and each group's requires — and walk it recursively, opening a dependent's span before its dependencies so it temporally encloses them.
• Roots parent directly to doctor run; a dependency shared by more than one group still runs exactly once, nested under whichever dependent reaches it first.
• Fixes two gaps a review of this change surfaced: a requires cycle (unvalidated anywhere today) previously left every member of the cycle unreachable from any root and silently dropped it from the run while still reporting success; a group missing from group_actions would have orphaned its entire dependency subtree instead of just itself.

Test plan

• cargo test — 205/205 passing, including 12 new tests (RunGraph unit tests, diamond/multi-root/fail-fast/cycle/missing-container behavior tests, and two span-nesting tests using a purpose-built tracing capture layer)
• cargo clippy -- -D warnings and cargo fmt --all -- --check clean
• Negative-control checks: reverted each of the four fixes individually and confirmed its dedicated test fails exactly as expected
• Stress-tested the runner test module for flakiness (15+ repeated runs, all green)
• Manually verified against local Jaeger using the issue's own MVCE — confirmed group mvce is the parent of group child-group and temporally encloses it (217484–217595 vs. 217504–217544)

Downloaded the MVCE from #172 and ran it against jaeger.

🤖 Generated with Claude Code