Feature Request: Native SWMR support for concurrent read/write

[14NGiestas]

Following my previous contributions (implementing iterate and visit), I've identified that h5fortran currently lacks high-level abstractions for SWMR (Single Writer Multiple Reader).
In scenarios where a simulation (writer) and a visualizer (reader, e.g., via h5py) access the same file, the default HDF5 locking mechanism often leads to crashes or file access errors. While this can be solved using low-level use hdf5 calls, it breaks the high-level philosophy of this library.
I am currently facing this exact issue: my visualization script (running in parallel with the main simulation) frequently crashes because of file locking/metadata contention. Providing a high-level SWMR interface would solve this real-world stability problem for users doing live data monitoring.
I'm opening this issue to keep track of the requirements for native SWMR support. Key components would include:

• File Access: Adding a swmr logical flag to %open() or %create(). This would internally handle H5Pset_libver_bounds(fapl, H5F_LIBVER_LATEST, H5F_LIBVER_LATEST) and the corresponding SWMR access flags.
• Writer Side: Exposing a %flush() method for datasets (wrapping h5dflush_f) to ensure metadata/data consistency is pushed to disk so readers can see updates.
• Reader Side: Implementing a %refresh() method (wrapping h5drefresh_f) to update the object's view without reopening the file.

Technical Context:
For the writer, the sequence usually requires:

1. Setting libver = latest in FAPL.
2. Creating the file and datasets.
3. Calling h5fstart_swmr_write_f.

For the reader:

1. Opening with H5F_ACC_RDONLY + H5F_ACC_SWMR_READ.
2. Periodic calls to h5drefresh_f.

I believe adding this would make h5fortran much more robust for real-time data monitoring. I'm documenting this here to track the feature progress and discuss the best API design before implementation.