Add function parameters to data flow analysis.

[cgyurgyik]

Addresses #120 .

[sampsyo]

Seems possibly reasonable to literally generate "fake" code for the hallucinated block… but it could be even simpler, I imagine to instead do it as part of the algorithm? The tricky thing, I think, is that it might require a new "hook" into the DF framework, i.e., different analyses may want to treat argument differently. For example, an interval analysis will want to treat the parameters as unknown values. I'll stare at it sometime too!

[cgyurgyik]

Seems possibly reasonable to literally generate "fake" code for the hallucinated block… but it could be even simpler, I imagine to instead do it as part of the algorithm? The tricky thing, I think, is that it might require a new "hook" into the DF framework, i.e., different analyses may want to treat argument differently. For example, an interval analysis will want to treat the parameters as unknown values. I'll stare at it sometime too!

Yeah I was a bit narrow-minded when writing this (also, trying to keep the PR as small as possible), thinking only about a few DF algorithms such as reaching definitions and "defined" analysis.

One thing to note is that the analysis's init value is determined with no consideration for the bril program, which isn't always the case? For example, how would one cleanly implement alias analysis with the current framework, where we want to have mapping from each variable to the empty set for init. Perhaps there's something I'm missing here.