Also removes resolved TODOs and removes usage of target where not needed

- You can use Fortran arrays as inputs, which are internally converted to C ptrs
- You can input normal Fortran strings which are internally converted to C strs
- You don't have to pass the size as an argument
- Return arrays are now Fortran allocatables arrays
- Jagged arrays are inserted/output as a single array and an array with the lengths of each dimension as opposed to cptr, cptr, size_n
- Pairs are now a (1:2,n) array that is cast into an assumed shape array in C
- `intent` keywords have been added to most variable types
- Had to create a dummy modue `fsize_m` to prevent the intrinsic `size()` being shadowed by any potential variables being named `size`

NOTES:
1. All the memory allocations in C need to be freed with a C callback otherwise we are leaking memory
2.

TODO List:
1. Add intent(out) for oint, osize, odouble
2. Add optional arguments for ierr
3. Add more generic optional arguments see Python
4. Allow for already allocated arrays to be deallocated internally

Because of how type-bound procedures are defined
the functions have to be parsed twice. First to generate the
the types and the second time to implement the interface with the C API.

`NOPASS` is used to signify that the procedures are static and

This is just syntactic sugar since the Fortran Standard does not allow
for intent(out), intent(inout) and volatile to be used with value.

Use lists instead of strs to store the argument names and types
This is necesseary because a Fortran procedure/interface needs both
a dummy argument list and a variable declaration following the arg list.

This also make it easier to add argument `intent` and also provide an
interface for optional arguments.

Optional arguments could either be defined on the `interface` block
or create a module implementation which calls `present`.
I think in the former case the optional arguments would be handed to
the C backend so e.g. stderr will be reported through C not Fortran

Temorarily removes call to flog until we finish the wrappers

The assumed size arrays have been converted into using the
`dimension(*)` keyword in an attempt to remove some of the
burden of formatting the Fortran interface from within the
variable definitions.

Newline characters have also been moved into the module generation
part of the script.

Fortran moules are public by default so limiting the number of
unecessary variables is important as to not populate the global scope.

- Removes non-standard DEC$ directives
- Changes comments to F90 (!) from F77 (C)
- Adds semver major, minor, patch variables
- Changes semver version to a larger string
- Changes docstrings to be Doxygen abiding
- Changes module name from `gmsh_fortran` to `gmsh`
- Changes file extension from `.h` to `.f90`
- Changes indendation to 4-space (fixed for now)
- Fixes & being erroneously placed at the start of newlines vs EOL
- Reformats variable commas, spacings, etc. to be consistent
- Removes hardlimit for procedures names> 45 to be excluded from the API

TODO List:
1. Abide by `indent`, currently spacing manually set to 4spaces
2. Add optional arguments for things such as `gmshModelAddPhysicalGroup`
3. Change names _n _nn to more descriptive names
4. Simplify spacing between procedure types
5. Convert API names to be user friendly get, set, etc.

extrude curve mesh depending on geo.Source (like for surfaces and volumes) instead of assuming that the extrusion is always done from the start point of the curve

create ghost cells in the usual way in PartitionUsingThisSplit() if they are not explicitly specified

OpenMP forbids leaving block via exception: catch exceptions inside the block, and throw the last error outside the block (cf. #1941)

use gmsh.view.option.*(tag, ...) instead of gmsh.option.*("View[index]...", ...) in tutorials (cf. #1649)

Could -> Could not

See merge request gmsh/gmsh!469