Preserve instance_id and execution_id in OrchestrationFailed events

[pinodeca]

Summary

HistoryManager::append_failed hardcoded placeholder identifiers ("" for instance_id and 0 for execution_id) that callers never overwrote. As a result, every terminal OrchestrationFailed event serialized an empty instance_id and execution_id = 0 into its payload, breaking payload-only correlation for telemetry/diagnostic consumers.

This affected more call sites than just the cancellation path. All four were fixed:

1. TurnResult::Failed arm — src/runtime/execution.rs
2. TurnResult::Cancelled arm — src/runtime/execution.rs
3. Ack-failure infrastructure error path — src/runtime/dispatchers/orchestration.rs
4. Poison-message failure path — src/runtime/dispatchers/orchestration.rs

Changes

• append_failed now takes instance_id: &str, execution_id: u64, matching the sibling Completed/ContinueAsNew terminal paths, and drops the placeholder comments.
• All four call sites now thread the in-scope instance/execution_id.
• Added a regression assertion to the cancellation sample (tests/e2e_samples.rs) verifying the terminal OrchestrationFailed event carries the real instance_id and a non-zero execution_id.

Compatibility

Backward compatible — no event schema change, only correcting values that were always meant to be populated. No rolling-upgrade concern. Existing already-persisted failed events are historical and not rewritten.

Testing

• All 70 cancellation tests pass.
• Updated cancellation sample passes.
• cargo clippy --all-targets --all-features is clean.

Fixes #35