Jinja seems to be taking longer than it should be

[bradley-solliday-skydio]

This can be seen when generating code in the python optimizer (though I imagine it can be seen just about anywhere Codegen objects are used to generate functions).

Perhaps something can be done to speed it up.

Here is a profile of symforce_py_optimizer_test.py, modified to generate 1000 distinct priors:

[bradley-solliday-skydio]

This has been sped up. Also, now only a tiny fraction of the time is spent in jinja functions (and instead almost all the time is spent in codegen functions):