Store scale in Column_definition::decimals rather than Column_definition::pack_flag

In order to simplify the unification of these Type_handler methods:

   virtual Field *make_table_field(MEM_ROOT *root,

                                  const LEX_CSTRING *name,

                                  const Record_addr &addr,

                                  const Type_all_attributes &attr,

                                  TABLE *table) const= 0;

and

  virtual Field *

  make_table_field_from_def(TABLE_SHARE *share,

                            MEM_ROOT *mem_root,

                            const LEX_CSTRING *name,

                            const Record_addr &addr,

                            const Bit_addr &bit,

                            const Column_definition_attributes *attr,

                            uint32 flags) const= 0;

let's do the following:

• Move Column_definition::decimals to the parent class Column_definition_attributes
• When loading a column definition from a FRM file, instead of mixing numeric scale to Column_definition_attributes::pack_flag, store it to Column_definition_attributes::decimals. This will be done in virtual implementation of Type_handler_xxx::Column_definition_attributes_frm_unpack().
• Mix (decimals << FIELDFLAG_DEC_SHIFT) to pack_flag only before storing it to the FRM. This will be done in virtual implementation of Type_handler_xxx::Column_definition_attributes_frm_pack().

As a result:

• Column_definition_attributes::decimals will always be up-to-date.
• Column_definition_attributes::pack_flag will never contain the decimal scale