fix(tests): remove ambiguity from managed execution checks

[moonyue-w]

Summary

Managed E2E verification could fail when the Agent and Skill both used sdk-proof-* names: the model confused the two resources and did not read the requested skill or memory. Use distinct sdk-agent-* and sdk-skill-* names, and supply the exact mounted file paths in the test prompt.

Before session cleanup, a failed turn now logs a bounded, sanitized tail of observed public events, including tool inputs, text results and stop reasons. Add offline coverage for diagnostic redaction and capacity limits, and document how to retain the test output.

Why

Make the requested resources unambiguous while preserving the existing marker, tool-use and completion assertions. Retain useful evidence after cleanup deletes the session, without retaining reasoning blocks, binary content, authentication headers, configured PAT values or URLs.

Verification

• make lint
• make build
• make test
• make docs-check
• go build ./examples/...
• Account-backed integration tests run, or not required

Offline checks passed with Go 1.26.4 against a clean archive of the base commit plus this patch. Untracked, ignored scratch tests in the local workspace are not included in that archive.

The targeted CN Managed run, using PAT authentication and model auto, passed all three subtests without skips: completion/SSE, file/environment, and skill/memory. Resource cleanup completed. Global and the full internal CI pipeline are not covered by this targeted run.

go test -tags live -count=1 -p=1 -parallel=1 -timeout=10m -json \
  -run '^TestManagedExecutionE2ELive$' ./managed
PASS — 3 subtests, 40.24s including setup and cleanup

Impact

• Public API or behavior changed
• User-facing changes added to CHANGELOG.md under Unreleased, or not applicable (explain below)
• Generated API documentation updated
• Forward/Managed contract fixtures updated
• Python and TypeScript parity considered
• Migration notes added for a breaking change
• No credentials, .env.live, logs, or generated test output committed

This changes Go test scaffolding and contributor guidance only. Public APIs, SDK behavior, fixtures and cleanup semantics are unchanged, so changelog entries, generated documentation and migration notes are not applicable. Python and TypeScript are outside the scope of this Go-specific failing test.