Suboptimal az path is chosen in some cases

The function recur is intended to find the best az values (taking into account branch options and sun avoidance) at which to run each operation.

I suspect this function isn't implemented correctly. The example in https://github.com/simonsobs/scheduler/issues/73 shows one consequence of the implementation -- the scheduler injects a move_to(az=52.0) unnecessarily; I think it could have just stayed at az=218 (or so) to do the bias_step. This happens because of some interaction with sun avoidance -- but I am finding it difficult to tell what the purpose of the function is, precisely.

I hereby request, for this specific az path optimization:

• unit tests (including a test that exposes the failure observed in #73, before it gets solved)
• clearer docstrings that state what the purpose of the function is (i.e. the problem it is solving / the metric it is using to optimize) and thus what the criteria for a successful test would be

Start with recur in src/schedlib/policies/stages/build_op.py around line 688, then review issue #73's example and the branch-option context in issue #74. Reproduce the unnecessary move_to(az=52.0) behavior and determine the function's intended optimization metric. Add unit tests, including a regression test for #73, and clarify the docstring so the success criteria are explicit.