cli: unwrap ExceptionGroup in MCP transport errors; status reports total_documents 0

[RichardHightower]

Summary

Two small CLI defects from the v10.7.0 integration test on 2026-09-19.

1. MCP transport errors hide the cause

Any MCP-side error reaches the user as the asyncio wrapper text, not the server's message:

$ agent-brain --transport mcp prompt explain-architecture
Error invoking prompt 'explain-architecture': unhandled errors in a TaskGroup (1 sub-exception)

$ agent-brain --transport mcp resources read "graph-entity://Symbol/OutOfStock"
Error reading graph-entity://Symbol/OutOfStock: unhandled errors in a TaskGroup (1 sub-exception)

The first is a missing required argument (--arg folder=...). The second is an invalid entity type. Both messages exist on the server side. The CLI should unwrap ExceptionGroup and print the innermost exception message (and, for a missing prompt argument, list the required arguments).

2. agent-brain status reports total_documents: 0 after a successful index

"indexing": {"total_documents": 0, "total_chunks": 13, ...}

The job log says 11 docs, 13 chunks. Either wire total_documents to the manifest file count or drop the field.

[RichardHightower]

Fixed in #257.

1. CLI unwraps ExceptionGroup / anyio TaskGroup wrappers so MCP errors (missing prompt argument, invalid entity type) print the server message instead of unhandled errors in a TaskGroup.
2. status.total_documents prefers the persisted manifest file count, so it survives the in-memory IndexingState reset after a successful index.