Specify behavior for empty or duplicate column names

[krlmlr]

Supported e.g. by RSQLite and MySQL. Need to assign a unique name, can perhaps borrow from tibble::repair_names().

[krlmlr]

odbc seems to return a data frame with an empty column name for SELECT 1, CC @jimhester.

[jimhester]

I actually prefer not assigning an arbitrary name for this, if someone writes the SQL without a name presumably it is for a reason and it seems strange to bring the data into R with one.

[krlmlr]

Many DBMS drivers will do this for you anyway. We could as well avoid specifying anything here, and leave it to the caller to repair column names.

@hadley @hannesmuehleisen: Any preference?

[hannes]

MonetDB(Lite) will assign a column name if none is specified:

> str(dbGetQuery(con, "select 1, 1"))
'data.frame':	1 obs. of  2 variables:
 $ L2: int 1
 $ L4: int 1

[hadley]

I think we should require that the backend provides a valid data frame - i.e. no empty, NA, or duplicated column names.

[krlmlr]

Also need to make sure that dbColumnInfo() is consistent with that dbFetch() produces.