Reduce code duplication and apply some good practices

[JsLth]

Some unorganized changes related to 1. refactoring, 2. increasing error robustness, 3. improving user feedback, 4. improving user friendliness, 5. improving understandability, some of which are arguable and/or subjective. The changes are not complete the pull request may be regarded as a draft, but I wanted to get this out before the Jour Fixe.

Key changes:

• Switched from manual object dispatch (point_*, poly_*, grid_*) to S3 class dispatch (closes Breakdown of point_*, poly_* and grid_* #16)
• Added a hash-based cross-session caching system (closes Feature: caching #10)
• Removed dependencies to dplyr, lubridate and tidyr (closes Reduce number of hard dependencies #3)
• Added an integrated way to handle API keys using set_api_key() (closes Show better user info when API keys are missing #8, questioning)
• Switched from progressr to traditional cli for more convenient info, warning and error messages (closes Leave use of with_progress to users #5)
• Added a function to stream ECMWF downloads and retrieve progress information (Progress information for the Shiny app #14, questioning usefulness)
• Added a way to discern server-side preparations and file downloads
• Changed term "focal" to either "toi" or "observation", depending on context
• Reduced code duplication by adding a couple of helper functions, particularly for preprocessing and raster extraction (not yet satisfied with this) (Be DRYer #7)
• Added a basis for a step-wise approach (.make_lnk) (Split linking functions #11, questioning)
• Avoid re-writing downloaded raster files if not necessary for parallelization (closes Only write netCDF files internally if parallel = TRUE #19)
• Added hard dependency for ncdf4 (closes Add a hard dependency to ncdf4 #2)
• Changed default download path to tempdir() without caching or tools::R_user_dir() with caching (closes Don't write in user files without explicit confirmation #4)
• Avoid changing input dataset, e.g. through reprojections (Redundant reprojections in grid_link_* #15)
• Some linting (e.g., sapply -> vapply)
• Split functions and R files in a functionally homogenous way
• Remove internal string to date conversion (reducing number of arguments)
• Set a default value for the date column and reduce necessary arguments by 1
• Set .data as the first argument to allow for piping
• Remove general imports and load specific functions using :: instead
• Add a verbose argument and a function info() that is suppressed if verbose = FALSE
• Probably a few more important things that I forgot