bug: propagate TaskExecutor shutdown failures

[u70b3]

Bug

TaskExecutor::shutdown_all logs handler cleanup failures and task panics but always returns Ok(()). Since ShardWriter::close merges that result into its close result, a failed task shutdown can be reported to callers as a successful close.

Expected behavior

• Preserve and return the first handler error or task panic.
• Continue joining every remaining task so shutdown is still complete.
• Keep the existing diagnostic logging for every failed task.

Reproduction

Register a message handler whose cleanup returns an intentional error, then call shutdown_all. The current implementation logs the error and returns Ok(()) instead of surfacing it.

Acceptance criteria

• shutdown_all returns an error when a handler cleanup returns an error.
• shutdown_all returns an error when a handler task panics.
• A failure does not prevent the executor from joining the remaining tasks.
• Existing successful-shutdown tests explicitly assert Ok(()) instead of discarding the result.
• ShardWriter::close continues to include task-shutdown failures in its staged close result.

Scope

This issue covers task-executor shutdown error propagation only. Other MemWAL close, durability, fencing, and index-validation behavior is out of scope.

[u70b3]

I am working on this and will submit a focused fix with regression coverage.