If you want to do this through editing you can find a simple method here. Also from this article: "There is a much simpler solution. If you run the Union tool on the feature class in question (so it is the only feature class in the list) and uncheck the 'gaps allowed' checkbox, it will fill in the gaps/donut holes which you can then query simply by the missing attributes."
You can also use the Merge or Append tools in ArcGIS to combine those polygons. The first creates a new output dataset and the second adds to an existing dataset. If you go this route I would recommend merging the polygons and then performing a union (which can eliminate any gaps). This may be your best bet since you can later add any nested islands.
MySQL should be following the WKT specification that was detailed by the Open Geospatial Consortium's Simple Feature Access - Part 1: Common Architecture.
The text you have is not valid WKT, and no GIS software will accept it (generally it will raise a parse error). Commas are used to separate coordinates and spaces between components of each coordinate. Furthermore with polygons, each linear ring needs to be closed, such that the first and last coordinate are repeated. Lastly, with Multi- shapes, you need an additional nesting of parentheses to consider 1 or more polygons.
In all, the valid WKT approximation of your attempt is:
MULTIPOLYGON (((0 0, 3 3, 0 3, 0 0), (1 1, 1 2, 2 2, 2 1, 1 1)))
Which is technically an invalid geometry, as it has a self-intersection at point (1 1). A valid polygon would have this interior ring completely contained within the outer ring, and not overlapping any other interior rings (if present).
If you want to get more familiar with WKT structures, I find JTS TestBuilder a really good program to visually interact with geometries.
Best Answer
This is fairly simple and you can do it all natively in QGIS (i.e.without exporting to PostGIS).
Step 2 is the key here. Step two creates a set of overlapping polygons, one of which represents your landmass with no holes and the others are your lakes. Step three kills the landmass polygon leaving just the lakes.
I vaguely remember a thread where somebody was asking how to de-doughnut polygons. You could use this same approach but instead of differencing your result with the original, simply dissolve the result, or do a spatial join and then select by attribute if dissolve is too crude... anyway, that's getting off-topic.