cnmatrix solver crashes depending on the finidat file, as well as in an IHist simulation

Summary:

cnmatrix may crash when given a finidat file from a different simulation, even when using init_interp.
cnmatrix may also crash in transient simulations.

Details: I have encountered three examples this far (see #640):

I2000 test using the original ctsm5.2 f19 finidat file:
- I generated this finidat from the 5.1 file by running ctsm5.2 with init_interp for one step.
- I found that the patch with the error had a crop type in the ctsm5.2 fsurdat file that was not present in the same grid cell in the ctsm5.1 fsurdat file. This indicates that init_interp did not obtain satisfactory (or any?) data for that patch.
- I ran a full new spin-up (https://github.com/NCAR/LMWG_dev/issues/58) and the I2000 test passed, so I have replaced the default finidat for cases such as this test.
NEON-NIWO single-point test with (ironically) the new ctsm5.2 f19 finidat file described in (1):
- I reran the test with the orig. ctsm5.2 f19 finidat file and it passed.
- The problem may be similar to (1)
- In this case we decided to replace the failing test with a NEON-HARV test, which is a better site to support than NIWO anyway according to @wwieder. This one works with the new ctsm5.2 f19 finidat file described in (1).
To spin up the model (see 1), I tried running all the phases (except AD) with cnmatrix; however, cnmatrix crashed in the historical simulation early in year 1901. The problem may be similar to (1).

Recommendations:

It seems safest to perform a unique spin-up for each single-point site and for each model resolution.
Also cnmatrix may be most useful in SASU mode because it results in much faster model spin-up.
Transient simulations with cnmatrix may not work.

CTSM version you are using: I'm using the cnmatrix branch (#640). The PR's notes show when the above issues emerged.

Configurations affected: cnmatrix

ESCOMP / CTSM