Mosart provides invalid starts/counts to scorpio

[dqwu]

When testing scorpio (PIO2) with latest E3SM on anlworkstation, 19 e3sm_developer tests failed:
https://my.cdash.org/viewTest.php?onlyfailed&buildid=1697655

Most of these tests did pass with scorpio_classic (PIO1) using latest E3SM, or with scorpio using a much older revision of E3SM.

One of the failed tests (ERS.ne11_oQU240.I20TRCLM45) shows the following error message:

Error termination. Backtrace:
At line 555 of file components/mosart/src/riverroute/RtmHistFile.F90
Fortran runtime error: Index '0' of dimension 1 of array 'rtmptr' below lower bound of 1

The affected line is:
field => rtmptr(hpindex)%ptr

By checking the restart file, we can find a lot of zero values for hpindex

int hpindex(max_nflds) ;
        hpindex:long_name = "History pointer index" ;
        hpindex:units = "unitless" ;
...
hpindex = 12, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
   0, 0 ;

The expected non-zero values (from passed test with scorpio_classic) are:

hpindex = 12, 11, 3, 2, 1, 4, 28, 27, 26, 25, 24, 23, 22, 21, 20, 19, 6, 5,
   8, 7, 14, 13, 10, 9 ;

We can confirm that the failure is caused by PR #3012 with git bisect. More specifically, a previous fix to components/mosart/src/riverroute/RtmIO.F90 (part of PR #2772 that fixes issue #2778) was accidentally reverted.

Without that fix, the starts/counts for puts where the vars (1d/2d/3d) had no time dimensions were wrong, i.e., always had an extra start/count corresponding to the time dimension.

When testing with scorpio these bugs showed up. Note, with scorpio_classic these bugs did not show up (inadvertently, since Fortran arrays always include the size).

We should reapply that reverted fix, and verify that it works for both scorpio_classic and scorpio.