BUG: regression in read_csv parser handling of usecols GH5766

GH4406 fixed one bug and introduced (exposed?) this corner case
where the loop over columns continues even after all usecols
have been fulfilled, causing an out-of-bound lookup in names.

Author: y-p

pandas/parser.pyx

@@ -898,6 +898,9 @@ cdef class TextReader:
             if i < self.leading_cols:
                 # Pass through leading columns always
                 name = i
+            elif self.usecols and nused == len(self.usecols):
+                # Once we've gathered all requested columns, stop. GH5766
+                break
             else:
                 name = self._get_column_name(i, nused)
                 if self.has_usecols and not (i in self.usecols or