Enhance H5 I/O

[dimitrivlachos]

Create a generic H5 writer — along with unit tests — that can take
generic data and write it to a file. As well as adding more robust error
handling to the reader.

• Create new subdirectory h5 in include/dx2 to logically group h5
  related code.
• Create new header file dx2/h5/h5write.hpp to define the writer.
• Move h5read_processed into dx2/h5 and add error handling.
• Create new test file test/test_write_h5_array.cxx for validation.
• Modify CMakeLists.txt to include the new header file, tests and
  changes to allow this to be included by other projects.

[dimitrivlachos]

@ndevenish I probably shouldn't keep all of this commented. Should I? Or is it something we can come back to later when we're going to need this?

[dimitrivlachos]

The decision is we will revisit this later when we need it as the repo is still in an early state

[jbeilstenedmands]

Thanks for adding this. This all looks and works fine, I'm not sure we need to fix the cmake installation as part of this work, better to address that separately in my opinion.
Just to note that when we use these functions to write multiple data arrays to a single h5 file, if we want the file to be readable as a dials reflection table, we will need to ensure that the arrays in a group all have the same length (in the leading/trailing? dimension for multi-dimensional vectors) - but I think this should be enforced by the reflection table class using these functions, rather than checking in these functions on each call to write.

[dimitrivlachos]

Thanks for adding this. This all looks and works fine, I'm not sure we need to fix the cmake installation as part of this work, better to address that separately in my opinion.

I agree. I think this will be fine to merge as-is — especially in these early stages — and we can come back and fix this as needed.

Just to note that when we use these functions to write multiple data arrays to a single h5 file, if we want the file to be readable as a dials reflection table, we will need to ensure that the arrays in a group all have the same length (in the leading/trailing? dimension for multi-dimensional vectors) - but I think this should be enforced by the reflection table class using these functions, rather than checking in these functions on each call to write.

Correct @jbeilstenedmands. As it stands, this implementation of write_data_to_h5_file naïvely writes any array of data, anywhere you ask it to. This is intentional, as the act of ensuring array length matching is not intrinsic to writing data to an H5 file. It is conceivable that write_data_to_h5_file could be used on its own, thus ensuring array length integrity across datasets in a reflection table will be the responsibility of the reflection table class and not the h5write implementation. 😄

[ndevenish]

After discussion we want to make some changes going on but this is fine to go in now for cross-development purposes.