Random-access for variable-based encoding (i.e. use SetStepSelection for ADIOS2 steps)

[franzpoeschel]

TODO:

• MPI-aware preparsing
• Maybe through out MPI-awareness of preparsing again

• Edge case: What if a particle species completely vanishes after some iteration, making it appear as an empty, i.e. constant component -> unsupported case, backend will throw

• Better testing
• Documentation, examples
• Replace group-based encoding with variable-based encoding more thoroughly for ADIOS2 (maybe follow-up)
• Can we make variable-based encoding the default output for ADIOS2? (maybe follow-up for easier reversibility)
• Check if the engine supports this mode ---> would include HDF5 engine, but it has no first-class support, so setting group-based encoding manually is fine
• Check if the datasets have the same number of steps as there are snapshots
• Add sth like "UnselectStepSelection" for reading /rankTable

• follow-up: maybe pre-parse the entire thing in linear encoding and only then switch to random-access?
• follow-up: Preparsing also in READ_LINEAR (i.e. eliminate duplicate Iterations better than now)

• For defaulting v-encoding: Do attributes such as dt or time make sense here at all? ---> Need to preparse those, too. Check all modifiable attributes upon first READ_LINEAR open. --> follow-up, print warning for now
• Maybe use openPMD 2.0 to introduce a more breaking default for v-based encoding in ADIOS2? --->follow-up
• Merge ADIOS2: Require v2.9 as minimum #1711 first

[ax3l]

#1711 merged, ready for rebase :)

[franzpoeschel]

#1711 merged, ready for rebase :)

done ;)

[franzpoeschel]

Note to self: Benchmark this on a Series with many Iterations, compare with group/file encoding and with variable encoding in linear access

[ax3l]

please include <optional> for stability in this file

[franzpoeschel]

Done, I've also removed an unused <map> import