Take std::filesystem::path where opm-common opens a file

[GitPaean]

1. Takes std::filesystem::path where opm-common opens a file — the EclIO readers and writers, Parser::parseFile(), and the keyword generator's file helpers — so a caller holding a path passes it as is. std::string still converts implicitly: no call site changes, and every consumer in opm-common, opm-grid and opm-simulators builds unchanged.
2. Removes the conversions the library was doing to itself (EclFile(rstFile.string()), EclOutput(smryDataFile.generic_string(), …), EInit(initFileName.generic_string())) and retires the parser's dataFileName[0] == '/' in favour of has_root_directory(), the same test expressed on the path rather than on its first byte.
3. No functional change: the names simply travel as the type the standard library opens files with, and nothing decides anything by inspecting their characters any more; the full opm-common test suite passes as before.

[GitPaean]

jenkins build this please

[GitPaean]

I consider this is a small improvements arguably, I would like to get some input before continue further. @bska @atgeirr @akva2

[akva2]

Makes sense to me. First commit would benefit from a split (path and fmt parts)

[bska]

I concur with @akva2: Taking path instead of string (or string_view) in those places that actually treat the string as a path name is the correct thing to do. I'd also prefer that fmt-related changes be split into separate commits, possibly even separate PRs.