Hold "go-describeTable" to codegen

[weiguoz]

Is your feature request related to a problem? Please describe.
In SQLFlow, we use "describeTable" to verify fields, read field type and gen python code. Such "describeTable" is read from database engine by database/sql driver. We mark it as "go-describeTable".

Meanwhile, we use db api to read "describeTable" and access database. We mark it as "python-describeTable".

Always, "python-describeTable" equals to "go-describeTable", but differ to sqlflow parser(sql.y). And each database has it's own "describeTable", as talked in "Add hive db-api in codegen".

statement:

SELECT *   
FROM iris.train 
TRAIN DNNClassifier
WITH n_classes = 3, hidden_units = [10, 20]
COLUMN train.sepal_length,train.sepal_width,train.petal_length,train.petal_width
LABEL class
INTO sqlflow_models.my_dnn_model;

• hive

field	type name
train.sepal_length	FLOAT_TYPE
train.petal_length	FLOAT_TYPE

Field name comes with a table name as prefix

• mysql

field	type name
sepal_length	FLOAT
petal_length	FLOAT

---

Describe the solution you'd like
If "go-describeTable" used to verify merely, the database description may be lost in codegen. The python code complains field not found. To avoid this, we should use sql.y's description to verify, and hold "go-describeTable" to codegen.