add mo_dag based on daglib by Jacob Williams

This MR adds a Directed Acyclic Graph implementation motivated by the routing module of mHM. All other changes beside mo_dag are related to the new routing routines.

• add mo_dag: Directed Acyclic Graph implementation

  • derived from https://github.com/jacobwilliams/daglib
  • added dependents array to nodes
  • simple graph traverse handler
  • level sorting based on Kahn's algorithm
  • order_t type to store level based ordering

• mo_datetime updates

  • add <step>_start, next_new_<step> and previous_new_<step> where step can be year/month/week/day/hour/minute
  • add add and sub for in-place addition and subtraction of timedelta
  • faster date/time comparison (no intermediate type instances)

• mo_grid updates

  • added dist_latlon, is_lat_coord and is_lon_coord to public routines
  • added id_matrix (full matrix of cell ids), cell_id (cell id from indices) and closest_cell_id (cell id closest to given coordinates) to grid class
  • better aux-coord finding when creating grid from netcdf

• mo_grid_io updates

  • add t_bounds to input_dataset to store values of the consecutive time bounds
  • faster time-stamp comparison
  • add time_index method to input_dataset to find index containing the given time-stamp (start time excluding, end time including in the bounds)
  • add grid_init_var to input_dataset%init to initialize grid directly from a given variable (shortcut)
  • add coordinates to output variables when writing (was missing before)

• mo_grid_scaler updates

  • renamed regridder to scaler_t to be consistent with naming
  • added no_scaling mode when source and target grid have the same resolution (only do flipping and masking if required)
  • fixed 2d downscale methods (where using wrong grid for indices in data before)

• mo_io updates

  • use plain read(u,*) in loadtxt to read floats

• mo_utils updates

  • add optval as helper function for most data types to set default values from optional values (internal_val = optval(val, default))
  • add support for i8 integers in swap
  • use i8 integers for indices in flip and flipped to support big arrays

• CI/CMake updates

  • added language C to CMake project to catch error when NetCDF is linked against HDF5 (e.g. libnetcdf 4.9.3 on conda forge (Pull Request)) and triggers the FindHDF5 script of CMake which tries to compile a test C program
  • cleaned up compile flags in CMake
  • added explicit support for IntelLLVM
  • add debug jobs for all intel compilers in CI