use generated templates to avoid code duplication

Since each rank needs a subroutine (and select rank wouldn't help much for this) we use generated code templates for this (via CMake or Meson). This increases code quality in general.

avoided the name %rank for possible confusion with intrinsic rank()

Noted that GCC 10.2.0 is making spurious warning over intrinsic :: rank that is needed to workaround an Intel compiler bug.

BUGFIX, features

• fix bug where finalize() closed ALL files instead of just one file.
• add hdf5_close() procedure, good to close at end of program to ensure no dangling files remain
• add flush() method, useful for long-running programs where periodic flushes help preserve data thus far if program crashes (run out of HPC time allotment for example)
• add is_open property, mostly used internally to guard against users calling methods on closed files, or double-initializing an open handle.

simplify pathlib generation

internal bugfix

status=scratch, more careful access handling

• status='scratch' much like in Fortran, expect you can specify an absolute directory. If not, Fortran uses system temp dirs it detects
• make access= parameter more clear and correct

add %ndims %rank method

These methods are equivalent

more robust config step, add "is_hdf5" file

• added is_hdf5(filename) function that returns .true. if a file is HDF5
• put stricter tests in CMake and Meson configure step to help user understand if there is a runtime HDF5 problem
• dedupe code by putting more into check() internal function

test robustness

test_deflate: allocate instead of stack memory to avoid random crashes of test. This only affects the test, not the library.

added h5exist(filename, datasetname) convenience function

robustness improvements

• use native cmake features instead of custom logic
• implicit none (type, external)

docs, cmake robustness

• docs: break up big readme into several linked files
• cmake: lint

add read slicing

added ability to slice arrays on read. This allows indexing into huge HDF5 disk variables and reading part of an array, for 1-D to 7-D.

Added HDF5 library build scripts, for those who need to build HDF5 themselves, under scripts/compile_hdf5.py