Add fix for pyodbc+mssql

[chinhngt]

• Handle Row object returned by pyodbc/msodbc
• Add tests

[chinhngt]

@marcusianlevine @xrmx Can you help to take a look? This should fix several issues reported earlier like:

#3934
#4210
#4211
#3075

[xrmx]

@chinhngt you missed to write what is the problem you are fixing. I suppose one between the tuple or the iterator breaks something else later in the chain. Or maybe both? I don't know the code but maybe it's the caller that must be fixed as an iterator may be a good idea to save memory.

[chinhngt]

@xrmx Currently when using Superset with pyodbc, the data pyodbc returns is in the form of [pyodbc.Row, pyodbc.Row, ...] instead of [(tuple), (tuple), ...] so SupersetDataFrame throws exception on wrong shape (it got 1 column instead of N columns). This change converts pyodbc.Row object to regular list before passing to SupersetDataFrame. I think BigQuery (BQEngineSpec in db_engine_specs.py) is doing similar thing :)

[xrmx]

@chinhngt thanks for the explanation. So you have to prepare data that pandas can digest for a DataFrame.
It looks like pyodbc always use Row classes as it implements the DB API, wouldn't it better to poke at the connection uri instead of looking at the data?

[chinhngt]

@xrmx I don't see sqla uri is available within db_engine_specs. Any hint?

[xrmx]

@chinhngt no clue, anyway it if works for you it's fine for me.

[chinhngt]

@xrmx Yes, it works for both Azure SQL and Azure Datawarehouse for me.

[bolkedebruin]

@mistercrunch as this PR was not rebased against master the license check did not kick in on the PR. It does have a new file and therefore master now fails building. To solve this have a PR with a license header on an existing file ;-)