Fix quadrature initialization in IDAS

[Steven-Roberts]

When IDAS takes its first time step, it calls the quadrature RHS' to fill in ida_phiQ[1] and ida_phiQS[1] vectors. When IDAS is reinitialized, it does not do this due to some logic errors. This caused a couple issues

• On the first step after reinitialization, the quadrature predictor is incorrect, so the quadrature error estimate tended to be a significant overestimate. Since the first time step is quite conservative, this usually didn't matter.
• When recomputing forward to fill in missing data between checkpoints, IDAS could produce very slightly different results than the original forward pass. This can cause a different number of time steps and eventually a segfault as found in a GridKit example (CC @pelesh)

[pelesh]

Thank you, @Steven-Roberts, for tracing this issue down and identifying the root cause. This has been showing up in GridKit tests nondeterministically. See here, for example. Changes in simulation configuration would make these issues come and go.

[pelesh]

I ran preliminary tests and this fix seems to address the issue we were seeing in GridKit. I need to run more tests to confirm.