BP5 min/max issue when using memory selection

[pnorbert]

Describe the bug
BP5 min/max is incorrect when a variable was written with SetMemorySelection. BP4 produces correct min/max.

To Reproduce

$ ctest -VV -R BPWriteMemSelReadVector.BPMemorySelectionSteps1D/.*.BP5.Serial
$ ./bin/bpls -la ./testing/adios2/engine/bp/bp5/BPSteps1D_1.bp i8
  int8_t          i8    3*{10} = -128 / 127
$ ctest -VV -R BPWriteMemSelReadVector.BPMemorySelectionSteps1D/.*.BP4.Serial
$ ./bin/bpls -la ./testing/adios2/engine/bp/bp4/BPSteps1D_1.bp i8
  int8_t          i8    3*{10} = 0 / 2

Expected behavior
Correct min/max of variable with every writing scenarios.

[pnorbert]

note: BP5 uses span for variables that has memory selection but then it does not compute minmax for spans. So fixing minmax for span will fix this bug. ( see BP5Writer.tcc:PutCommon():60-77

[eisenhauer]

I was just going to add that this was really a Span problem... But generally I'd like to discourage span with BP5 because with the write vector it's not a win. Wondering if we couldn't just get by with doing something different in BP5writer.tcc inside if (!variable.m_MemoryCount.empty()). But maybe the straightforward thing is just to fix min/max with span.

[pnorbert]

it seemed more complicated to mess with Marshal() to make it aware of mem-selection

[eisenhauer]

In this case, I was thinking about reordering the memselection bit so we copy first and then marshal without using a span, rather than messing with marshal. But that still leaves us without min/max for span.

[pnorbert]

triple copy?

[eisenhauer]

At most double, but that's bad enough. I'll have put this on the list to figure out.

[eisenhauer]

OK, I have a fix coded and can verify that it works with bpls. But I need to write some tests so it says working.

[eisenhauer]

Closed by #3306.