TODO items

[mgates3]

This is a loose collection of TODO items.

• Standardize order of layout, priority, queue. See PR Remove tile life infrastructure #151 for comments.
• Rename tbsmPivots => tbsm_pivots
• Use tag_XYZ where appropriate. E.g., listBcast in work_trsmA.
• tileRecv can generate an OpenMP task. Is this desired? Do algorithms need to wait for the task? See comment in Avoid allocating unneeded host tile #122 and Commits that didn't get pushed to #122 #161.
• Use tr_matrix in run_tests.py. See comment in Add nonuniform tile sizes, device distributions, and grid orders to even more testers (depends on #143) #157.
• Address "fix gemmA on device" comment in test_her{k,2k}.cc. See comment in Add nonuniform tile sizes, device distributions, and grid orders to even more testers (depends on #143) #157.
• Audit MPI tags to check for potential overflow; MPI only guarantees 32768; Intel and Cray 20-21 bits, not 32-bit values. See comment in Add nonuniform tile sizes, device distributions, and grid orders to even more testers (depends on #143) #157.
• Add C generics, Fortran interfaces. See comment below.
• Add uplo or diag accessors in the C and Fortran APIs. See note in Fix Options in C and Fortran APIs #152.
• Update c_api Python scripts to write blocks. See discussion in Fix Options in C and Fortran APIs #152.
• Add user-level example of setting entries on the GPU. See user forum email "Is it possible to access matrix elemnts within Cuda?"