address issue with max concurrent and work fetch

[davidpanderson]

Fixes #5749 (hopefully)

Max concurrent is a limit on jobs, not processor instances.
The work fetch logic made the erroneous implicit assumption
that all jobs use 1 CPU.
So e.g. if project has max concurrent 4,
and the client has two 2-CPU jobs,
it will think (if work buf is zero) that there's
no point in fetching more work.
But in fact the project could use 8 CPUs, so 4 are idle.

Fix: if a project has MC constraints,
then for each resource compute 'mc_max_could_use':
the max # of instances the project could use, given its MC constraints.
Use this to compute the project's shortfall,
and hence to decide whether to fetch work from it.

Note: the way mc_max_could_use is computed is crude;
it takes the max over all apps,
when it's possible that only one of them has a MC constraint.
This could result in limited over-fetching,
but that's preferable to under-fetching and starvation.

Sim: show app name in timeline

[AenBleidd]

@davidpanderson, are you sure this is the fix for #5743, not for #5749?

[davidpanderson]

Oops! Fixed.