public / sequelize

Major overhaul of the createTableQuery to bring it closer in line
with the mysql dialect. This closes a number of failed cases where
things were being created in a strange manner, as well as let's us
do away with SqlGenerator once and for all

These tests must be disabled since MSSQL simply doesn't allow
the updating of these types of fields. Cascading updates are supported
and should be covered by other tests.

Instead of trying to parse the error messages in the connection
manager, forward all errors to the user to deal with on an as-needed
basis

There are more than one error messages related to unique
constraints that we need to catch. This closes one more broken
test, in ancitipation of the crate table refactoring which closes
the other.

There is no need to use the mssql wrapped around tedious, and
switching to the lower level module allows better integration
as well as bypassing the mssql forced connection pooling

Code was merged into master to allow modification of the abstract
dialect to support mssql specific cases. This removes the custom
implementations

This is more of a stylistic change, as no functional behavior differs,
however it makes diagnostics much easier and is theoretically faster
than the previous select code.

MSSQL doesn't support RESTRICT foreign key constraints at all,
so there's no use running any of those tests

This is completely broken anyway, these tests are being used to
conditionally report back false positives for transaction support.

 - modify the returned sql for isolation level
 - remove some commented out code
 - cheat a little by reordering tests (look into this)

  - major deletion of code duplication
  - cleaned up code (removed duplicate documentation)
  - moved a lot of SqlGenerator sql back into QueryGenerator

  - missed a case for addLimitAndOffset for subqueries with a single model
  - formatResults should return undefined if there is no data
  - there are a few possible error messages in an invalid connection scenario

  - bring addIndexQuery in line with abstract dialect
  - use an improved method for showing indices with sp_helpindex
  - actually use handleShowIndexQuery (bringing us closer to the
    abstract dialect for formatResults)
  - temporarily disable tests for a FULLTEXT index

DataTypes.DATE was set to "DATETIME2", which has implications for internal
sequelize code. Instead we should be using a data type mapping, like postgres
support.

VIRTUAL attributes were accidentally being included in INSERT and UPDATE
statements as OUTPUT variables

Bulk insertions were refactored to prefer row construction for bulk
inserts that are not empty. The actual sql generation was moved back
into the query-generator, and removed from sql-generator. Also a number
of console.log and console.warn's were moved over to using sequelize's
internal logging mechanism

There a whole bunch wrong here, but mostly with sequelize. First and
foremost attributeToSQL (the convention) should be working with attributes,
however it's using dataTypes here because.. well every other dialect does
this too. addColumnQuery as defined in the abstract QueryGenerator class
says you should be using attributes, but everyone else uses (key, dataType).
This is a kludge to make ENUMs work for TSQL, since it needs to know the
field name to generate the SQL

The existing timestamp test checked the values within 50ms of each
other, since 50ms was a realistic latency for the existing test
framework (locally on travis with pg, mysql, etc). The mssql tests
generally take much longer, but this solution accounts for both and
future cases.

The MSSQL dialect implements its own sql generator, and was accidentally
adding an extra raw attribute for an aliased field