Clean up the representation of string types

[bocchino]

Now that we have the default string size available in the model, we should change the spec and implementation to (1) remove the “string with no size” type (that is, give every string type a size) and (2) store size in the Scala representation of the string type. This will simplify the typing rules, simplify the implementation, and improve the code generation, since we can use the numeric value of the default string size.

Related to #725.

[bocchino]

On further investigation, I think we should not make the changes exactly as described above, for the following reasons:

1. It turns out the string type with no size is useful, to represent the type of a string literal. If we don't have this type, then every string literal depends on FW_FIXED_LENGTH_STRING_SIZE. This seems unnecessary, since we don't care about the fixed length string size until the point that we need to generate code for a string variable.
2. This approach makes it more difficult to compute common types. In general we might not know the default string size yet at the point where we need to compute a common string type. Perhaps we could treat "no size" as a placeholder for "default size" and fill in the default size later, but this seems unnecessarily complicated. Again, we only care about these sizes at the point where we are code generating variables or serializing string values. So there seems to be no reason to compute these precise intermediate types.

I will revise the scope of this ticket as follows: go through the spec and the implementation and do some minor cleanup, to make sure they are consistent. However, I think the general approach we currently have (type of string literal is string, common type of string types is string) is the correct one.