feat: Umpire csv stats output

[MelReyCG]

This PR aims at adding a new MemoryStats component to add in Outputs to control where the Umpire report will be output (log by default, CSV optionally).

It also fixes the percentage value computation (which did not match what the description was telling, it was computed only for rank 0 rather than for all ranks).

Here is a generated CSV:

Cycle,Time,Umpire Memory Pool,Min rank mem bytes,Min rank mem %,Max rank mem bytes,Max rank mem %,Avg rank mem bytes,Avg rank mem %,Sum rank mem bytes,Sum rank mem %
0,0,HOST,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855
1,5000,HOST,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855
2,10000,HOST,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855
3,15000,HOST,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855
4,20000,HOST,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855
5,25000,HOST,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855,2913644,0.00035934855

The same CSV aligned for readability:

Cycle, Time , Umpire Memory Pool, Min rank mem bytes, Min rank mem %, Max rank mem bytes, Max rank mem %, Avg rank mem bytes, Avg rank mem %, Sum rank mem bytes, Sum rank mem %
    0,     0, HOST              ,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855
    1,  5000, HOST              ,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855
    2, 10000, HOST              ,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855
    3, 15000, HOST              ,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855
    4, 20000, HOST              ,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855
    5, 25000, HOST              ,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855,            2913644,  0.00035934855

Here is the new table output, which may be clearer:

----------------------------------------------------------------------------------------------------------------------------------
| Umpire Memory Pool | Min reserved rank memory | Max reserved rank memory | Avg reserved rank memory | Sum reserved rank memory |
|--------------------|--------------------------|--------------------------|--------------------------|--------------------------|
|                    |   bytes   |  over total  |   bytes   |  over total  |   bytes   |  over total  |   bytes   |  over total  |
|--------------------|-----------|--------------|-----------|--------------|-----------|--------------|-----------|--------------|
|               HOST |    2.8 MB |        0.0 % |    2.8 MB |        0.0 % |    2.8 MB |        0.0 % |    2.8 MB |        0.0 % |
----------------------------------------------------------------------------------------------------------------------------------

---

This PR also moves the parallelThreads attribute from OutputBase to SiloOutput because it was only used there. It didn't felt right to me to have a visible & documented attribute that had no effect, if I understood what it does correctly.

[arng40]

Since good() check the state of stream, why don't we check this condition every time (headerToStream & dataToStream)?

[MelReyCG]

I don't understand, as headerToStream() is based on toStream(), isn't good() verified correctly?

[arng40]

What I mean is why we can't get rid of bool isNewlyOpened and it will simplify the whole process.

[MelReyCG]

Is your suggestion to extract this code from toStream() directly in the TableCSVFormatter methods?
If not, can you show a short code snippet? I don't visualize what you mean.

[MelReyCG]

I got it, isNewlyOpened complexifies the code, and is not accurate in some scenarios. I'll remove it and edit the message.

[MelReyCG]

@corbett5 @wrtobin @CusiniM, can I ask you for a review?

[MelReyCG]

Ok, I think the PR is now ready, it will need a rebaseline because of the parallelThread moving.