Separate pragma parsing out of language parser and put it into model builder

[ywei]

The "#pragma" representation need to be separated from the language parser. The MiniMP language parser should just preserve the pragmas as strings in the AST. The model builder should parse these pragmas and construct the corresponding components from them. A pragma parser is needed to add. (using Antlr, for example)

[Stephen Siegel]

This is superseded by new front-end pragma parser.