fix: use base-2 mebibytes for memory requests based on LSF behavior - #9

The LSF documentation is very confused and contradictory regarding whether memory requests and
reporting are in base-2 or base-10. Here is an adaptation of my notes from a private chat regarding
the behavior in practice on the St. Jude HPC:

---

LSF allows the following suffixes:

```KB or K (kilobytes)
MB or M (megabytes)
GB or G (gigabytes)
TB or T (terabytes)
PB or P (petabytes)
EB or E (exabytes)
ZB or Z (zettabytes)
```

Here's an example. 16 gigabytes should be 16000 megabytes or 16000000 kilobytes, right?

```
LSB_JOB_REPORT_MAIL=N bsub -oo lsf.out -eo lsf.err -R "rusage[mem=16000000KB/job]" sleep 20
```

```
Job <274348028>, User <afoltzer>, Project <default>, Status <RUN>, Queue <stand
                     ard>, Job Priority <50>, Command <sleep 20>, Share group c
                     harged </afoltzer>, Esub <hpcf>
Thu Nov 13 20:34:14: Submitted from host <noderome104>, CWD <$HOME>, Output Fil
                     e (overwrite) <lsf.out>, Error File (overwrite) <lsf.err>,
                      Requested Resources <rusage[mem=15625.00/job]>;
Thu Nov 13 20:34:20: Started 1 Task(s) on Host(s) <noderome209>, Allocated 1 Sl
                     ot(s) on Host(s) <noderome209>, Execution Home </home/afol
                     tzer>, Execution CWD </home/afoltzer>;
Thu Nov 13 20:34:28: Resource usage collected.
                     MEM: 7 Mbytes;  SWAP: 0 Mbytes;  NTHREAD: 4
                     PGID: 3985845;  PIDs: 3985845 3985875 3985877

 MEMLIMIT SWAPLIMIT
   15.2 G    15.2 G

 MEMORY USAGE:
 MAX MEM: 8 Mbytes;  AVG MEM: 4 Mbytes; MEM Efficiency: 0.05%

 CPU USAGE:
 CPU PEAK: 0.00 ;  CPU PEAK DURATION: 0 second(s)
 CPU AVERAGE EFFICIENCY: 0.00% ;  CPU PEAK EFFICIENCY: 0.00%

 SCHEDULING PARAMETERS:
           r15s   r1m  r15m   ut      pg    io   ls    it    tmp    swp    mem
 loadSched   -    3.5    -     -       -     -    -     -  1000M     -      -
 loadStop    -     -     -     -       -     -    -     -     -      -      -

          waitpct     cWs   MBWps rgs01_load
 loadSched   10.0    40.0 20000.0      130.0
 loadStop      -       -       -          -

 RESOURCE REQUIREMENT DETAILS:
 Combined: select[(rhel8) && (type == any)] order[r15s:pg] rusage[mem=15625.00/
                     job] affinity[core(1)*1]
 Effective: select[((rhel8) && (type == any))] order[r15s:pg] rusage[mem=15625.
                     00/job] affinity[core(1)*1]
```

Note in particular the mem and swap limit of `15.2 G` and the fact that the rusage string now reads
`mem=15625.00/job` (the default unit on _our_ cluster but not all LSF clusters is `MB`). So, whoops!
I misremembered: sometimes it looks like you've been given less than what you asked for, but that's
because it's reporting in base 2!

Now let's try 16777216 kibibytes, which is 16384 mebibytes or 16 gibibytes.

```
LSB_JOB_REPORT_MAIL=N bsub -oo lsf.out -eo lsf.err -R "rusage[mem=16777216KB/job]" sleep 20
```

There are 16777216 kibibytes in 16384 mebibytes or 16 gibibytes.

```
Job <274346639>, User <afoltzer>, Project <default>, Status <RUN>, Queue <stand
                     ard>, Job Priority <50>, Command <sleep 20>, Share group c
                     harged </afoltzer>, Esub <hpcf>
Thu Nov 13 20:30:10: Submitted from host <noderome104>, CWD <$HOME>, Output Fil
                     e (overwrite) <lsf.out>, Error File (overwrite) <lsf.err>,
                      Requested Resources <rusage[mem=16384.00/job]>;
Thu Nov 13 20:30:27: Started 1 Task(s) on Host(s) <noderome140>, Allocated 1 Sl
                     ot(s) on Host(s) <noderome140>;

 MEMLIMIT SWAPLIMIT
     16 G      16 G

 MEMORY USAGE:
 MEM Efficiency: 0.00%

 CPU USAGE:
 CPU PEAK: 0.00 ;  CPU PEAK DURATION: 0 second(s)
 CPU AVERAGE EFFICIENCY: 0.00% ;  CPU PEAK EFFICIENCY: 0.00%

 SCHEDULING PARAMETERS:
           r15s   r1m  r15m   ut      pg    io   ls    it    tmp    swp    mem
 loadSched   -    3.5    -     -       -     -    -     -  1000M     -      -
 loadStop    -     -     -     -       -     -    -     -     -      -      -

          waitpct     cWs   MBWps rgs01_load
 loadSched   10.0    40.0 20000.0      130.0
 loadStop      -       -       -          -

 RESOURCE REQUIREMENT DETAILS:
 Combined: select[(rhel8) && (type == any)] order[r15s:pg] rusage[mem=16384.00/
                     job] affinity[core(1)*1]
 Effective: select[((rhel8) && (type == any))] order[r15s:pg] rusage[mem=16384.
                     00/job] affinity[core(1)*1]
```

16 gibibytes is 16384 mebibytes, so if we give a base 2 memory constraint in rusage we do in fact
get a base 2 amount of memory reported by the system. Even though all the docs use the base 10 unit
names and suffixes.