Implement estimation under discretionary optimal policy

Also add a test case.

For the time being, estimating parameters that appear in the discount factor is
not supported.

Ref. #1173

(cherry picked from commit 177c5c35)