I think you are on the right track.
- Having one document for each stock symbol will give you a good overview of all the symbols in the collection. And each document will have a fairly maintainable size.
- In my opinion, if you are even close to 16MB on a single document, the schema-design is far from good enough. It's not easily readable or maintainable. You also have to fetch a whole lot of data each time you want anything from the document.
- You mention "one docuement per symbol per day". To me that sounds like a sensible way to structure the data. Although i'm not familiar with the details in tick data from stocks, I supposed this will give you a good foundation for the schema design. You split it by each day, and can easily get all ticks for a given day/hour.
- Remember, there is no absolute solution to schema-design, as long as you think through it thoroughly. (there is definitely a right/wrong way though) ;)