[codex] Trace skill prompt injection paths

[alexsong-oai]

Summary

• Add tracing around core and extension skill selection, prompt reads, and instruction fragment rendering.
• Log host, executor, and orchestrator skill resource reads, including cache hits and read sizes.
• Keep this as a stacked PR on top of [codex] Emit implicit skill usage for support reads #29731 so the diff isolates the additional diagnostics cleanup.

Impact

This makes skill prompt injection easier to debug across host, executor, and orchestrator-backed skill sources without changing the intended user-visible behavior.

Validation

• git diff --check origin/codex/emit-implicit-skill-usage-events...HEAD

Full crate/workspace tests were not run in this PR creation pass.