SQL: add multi-index support to legacy mode

[jorisvandenbossche]

Closes #6881.

@danielballan Added multi-index support to legacy mode, wasn't actually not that difficult (all tests seems to pass). And at once also index_label kwarg is supported in legacy mode.

This just leaves the issue that this is actually an api change, so could possibly break code.

[danielballan]

That was quick. This seems like the best outcome from an assortment of confusing compromises, any of which could mean trouble for legacy code. Looks good to me.

I will proceed with #6796 once this is merged.

[jorisvandenbossche]

@jreback @hayd What do you think of the API change (before: no index was written, now index kwarg to write index which is set to True by default)?

[jorisvandenbossche]

Actually, we could also set index to False when using write_frame, and leave it as True for to_sql. to_sql is a new function, so for that the backwards incompatibility is not a problem.

(at this moment, write_frame is just passed in its entirety to to_sql)

[jorisvandenbossche]

@jreback @hayd OK with this? I added a commit that changes the default index to False for write_frame (behaviour of 0.13), for to_sql (new function) it is then index=True. As this is the most backwards compatible I suppose no objections?

[jreback]

looks ok

[jorisvandenbossche]

Ahum, I was actually completely wrong here... to_sql did already exist of course, it was added in 0.12. It was just not in the sql module but only as a method of DataFrame where it called directly write_frame, and it was not mentioned in the docs.
So my reasoning was not correct, and for to_sql this remains a backwards incompatible api change.

[danielballan]

Bummer. But I still think this is the lesser evil. Dropping the index was always a weird behavior, an indication that the SQL capability wasn't mature and should be used cautiously.