Added generalisation of png metadata handler

[WillTrojak]

This is in answer to #1155.

The intention is to generalise the handling of setting metadata in the PNG files. I am open to changes in the syntax that I've used or which domain is chosen to set the metadata from.

At the moment, the syntax that I've used is that for the first child domain, any child of the state node that has a child called keyword and data will create a png metadata keyword. ie

domain_0:
   state:
      domain_id: 0
      time: 
         keyword: 'Time'
         data: 0.1
domain_1:
   state:
      domain_id: 1
      comment: 
         keyword: 'Comment'
         data: 'some comment'

Will give the png metadata:

Time: 0.1

This seemed like the cleanest way of handling this.

[cyrush]

@WillTrojak good catch and nice suggestion!

They way ascent is currently plumbed, PNG comments are only used MPI task 0, (since task 0 writes the final pngs)

For the general case, it's possible that task 0 may have no domains at all. This is rare but we need to support it.

So we will need an extension to your code to find the first task that has this info, and then get that info to Task 0 (or all tasks).

That will require some MPI coordination. I can help implement this, but won't have time until later next week.

[FreddieWitherden]

For the general case, it's possible that task 0 may have no domains at all. This is rare but we need to support it.

This is quite likely in the PyFR use case as we often only pass Ascent a small subset of the total computational domain. As such it is not at all unusual for ranks to be excluded and have no data of their own (but they all make collective calls to Ascent).

[WillTrojak]

I added a simple test so that the first domain that has metadata is used. Let me know if you forsee any issues with this.

[cyrush]

Yes, this looks good. I think we should get this in now.

Sorry for the delay on the MPI version of the checks.
I can follow up with those changes later.