prio_notes

In the diagonal chare (factorDiagBlock):

- factoriDiagBlock is critical

- colMax is the target of a reduction, and lies on the critical path in
  the factorization

- doPivot is sent to both itself, and all the chares in its row and
  below. Urgency varies as follows: 

  * local invocation is immediately critical, as is any remote partner
    - stop waiting on a self send
  * invocation on the column below is critical before the next column
    can proceed
  * the row to the right and any partner chares need to hear before
    subsequent updates can happen
  * the sub-matrix below and to the right needs to hear before they
    can participate in factoring the next column panel or compute
    trailing updates
  * the sub-matrix below and to the left needs to hear before they can
    contribute their chunk of the solution

- sendPivotData is critical when it happens

- sendUSegment needs to be delivered and processed before the next
  reduction to colMax can start


In the off-diagonal chares

- doPivot/sendPivotData is interleaved with trailing updates, and can
  block execution of memory-relieving trailing updates

- recvL/recvU are trivial and can be necessary to relieve memory
  pressure, by scheduling a trailing update

- trailing updates need to happen before a block is part of the active
  panel, but should otherwise be delayed

- Above diagonal doPivot/sendPivotData need to happen before the
  portion of U can be computed. Very important at the left end,
  progressively less so as we move right. sendPivotData can be
  individually prioritized, since it's a point-to-point message

- processComputeU's priority depends on how much of the computation
  hinges on its completion. Above a soon-to-be-active diagonal block,
  important, further right less so.

- Below diagonal, the active sendPivotData calls are critical, and
  multicastRecvL is of varying importance depending on the
  recipient. LATER sendPivotData calls aren't needed until the solve.