Resource graph reinitialization with existing jobs leaves idata inconsistent

[milroy]

As @zekemorton discovered during test case development of PR #1053, reinitialization of the resource graph (i.e. here) leaves the resource graph idata in an inconsistent state. As an example:

$ ./resource-query -L ../../t/data/resource/jgfs/elastic/node-test.json -f jgf -S CA -P low
INFO: Loading a matcher: CA
resource-query> m allocate ../../t/data/resource/jobspecs/elastic/test-ma-node.yaml
[snip]
      ---------------core34[1:x]
      ---------------core35[1:x]
      ------------socket1[1:x]
      ---------node0[1:x]
      ------rack0[1:s]
      ---medium0[1:s]
INFO: =============================
INFO: JOBID=1
INFO: RESOURCES=ALLOCATED
INFO: SCHEDULED AT=Now
INFO: =============================
resource-query> attach ../../t/data/resource/jgfs/elastic/node-add-test.json
resource-query> cancel 1
rem_agfilter: planner_multi_rem_span returned -1.
medium0.
No such file or directory
ERROR: error encountered while removing job 1
resource-query> 

The error is caused by replacing a populated planner_multi_t with an empty one here:

flux-sched/resource/traversers/dfu_impl.cpp

Line 1137 in 021d180

(*m_graph)[u].idata.subplans[s] = p;

To solve this, we will likely need a way to partially remove a span from planner and planner_multi and update their resource counts.

[milroy]

Closing since this was resolved in PR #1061.