Mainly, C-casting of the result of a mem_root malloc or
calloc is replaced with actually calling the
constructor. Sometimes, however, the TABLE_LIST is part of a
larger chunk of memory. In these cases, that practice is
untouched and a placement-new is used instead.

MyISAM code still commits the above atrocities since we
don't want to touch it.

Fortunately, most members of TABLE_LIST already have default
initializers which keeps the constructor lithe and readable
even without the zero-initializations.

Sometimes TABLE_LIST objects are copied, and this has been
replaced with the obvious choice of copy-construction. This
revealed that MDL_key objects (member of MDL_request objects
which is a member of TABLE_LIST) forbade copy-construction
but were copied behind our backs using memcpy(). This
copying is now made visible by enabling the copy-constructor
and it's also possible to break in a debugger.

Another hack also got uncovered: dd::enum_table_type has a
fourth value of 0 which is invisible since it's a
side-effect of the zeroing-out. This is probably why the
enum explicitly starts on 1. This is fixed by adding the
enum value INVALID_TABLE.

Change-Id: Ic9d1f58dca36deba6a9be33fd5e2c2e8b3dcd49d