(The above problem only occurs with -T -- create a separate file for
each table / view.) This ChangeSet results in correct output of view-
information while omitting the information for the view's stand-in
table. The rationale is that with -T, the user is likely interested
in transferring part of a database, not the db in its entirety (that
would be difficult as replay order is obscure, the files being named
for the table/view they contain rather than getting a sequence number).


client/mysqldump.c:
  Added missing fclose(). Before, a view's stand-in table would get
  dumped in get_table_structure(), and the file would remain open.
  get_view_structure() would re-open the same file and write to it,
  resulting in garbage.  The way we handle it now, the table-struct
  gets closed, then the opening of the view-struct (same name)
  overwrites it. (The SQL for the view drop-if-exists the table,
  anyway.) If this were not desired and we wanted SQL for the views
  that contains the create for the stand-in table, we'd hand a mode
  to open_sql_file_for_table(), which would feature O_APPEND in
  get_view_structure(), but not in get_table_structure().
mysql-test/r/mysqldump.result:
  prove mysqldump -T (each item gets its own file) dumps views correctly
mysql-test/t/mysqldump.test:
  prove mysqldump -T (each item gets its own file) dumps views correctly