Curious error logic in completion--some

[Daniel Colascione]

Why does completion--some to raise an error only if one of its passed-in 
functions raises an error and all return nil? If we want to treat an 
error as "no completions", we should do that consistently or 
consistently propagate errors to callers. The way it is now, where the 
interpretation of an error from a completion table depends on what other 
completion tables are in use, seems odd.