Stop ceres problem from freeing loss function

[john-maidbot]

---

Basic Info

Info	Please fill out this column
Ticket(s) this addresses	N/A
Primary OS tested on	Ubuntu
Robotic platform tested on	gazebo simulation

---

Description of contribution in a few bullet points

• Stop ceres problem from freeing the loss function when we reset the solver. This is important because otherwise calling solver_->Reset() invalidates the loss_function_ pointer held by the solver_ (whereas I would expect the behavior of solver_->Reset() to be that the loss function is left unchanged). Open to alternative ways to address this problem, but this worked for my use case.
• It also seemed like the blocks_ object should be freed in the ~CeresSolver() function (since I don't think the problem takes ownership of this).
• I ran into this issue because in my use case, I need to cleanly reset the slam toolbox state multiple times without killing the process (funny enough you don't run into major issues until the second time you call solver_->Reset()). I also think this would affect people if they needed to call deserialize pose graph multiple times.

Description of documentation updates required from your changes

None that I know of.

---

Future work that may be required in bullet points

• I hope to make a future PR with the reset function if it makes sense for general use cases. atm it seems unique to my use case.

[SteveMacenski]

LGTM, thanks!