Improve performance of Python generation

[joshsh]

Now that the Python coder and supporting kernel modules are able to transform the entire Hydra kernel into Python, we need to improve their performance. Python generation is currently much slower than Haskell generation. This is partially explained by the extra inference phase during adaptation, but not completely. Additionally, special settings are required in order to prevent stack overflows during generation of particular modules like hydra.rewriting, probably because they are larger and more complex. We are going to be doing a lot of Python generation in the next few weeks, so addressing the performance issues sooner rather than later will save time in the long run.

[wisnesky]

making all computations in haskell strict is usually a good start

[joshsh]

That's doable, because Hydra does assume lazy evaluation. However, in this case I am thinking more in terms of eliminating unnecessary steps -- like cutting out that second inference phase by making all of the adapter rewriting steps type-preserving.