[misc] cleanups and avoid some unneeded Rcpp::String

[JanMarvin]

TL;DR: This PR lifts the burden of calling Rcpp::String() for the cc columns r, row_r and c_r, which are filled for every cell of data frame (A1, 1 A). This should bring some speedups.

Back in the day we were troubled by non unicode systems which were unable to convert unicode characters into their locale. The solution is to pass strings through Rcpp::String() which makes sure that the correct locale is picked up. Unfortunately this is quite slow in comparison to std::string() (IIRC every string has to be created in R, contrary to std::string() which does not have to be passed to R). Using Rcpp::String() is not only slow, it also worsens with more data. Ultimately, if Rcpp::String() is called a lot, it will slow things down.

Sadly we will have to remain using Rcpp::String(), for the foreseeable future, because non unicode locales wont go away anytime soon. Maybe another solution is to use enc2utf8() and enc2native(), similarly to how IIRC cpp11 does it, but this would require other code changes and I'm not entirely sure that we will see additional gains.

Still, we do not have to pass every cell of every cc column through Rcpp::String(). There are AFAIK only a few that can actually contain unicode strings and only the ones containing unicode characters, actually benefit from Rcpp::String():
is and maybe ph and in some rarer cases v (only in combination with str, otherwise it should always return integers or numerics) and f (formula using UTF8-strings). Hence, we should treat only these.

We have many more Rcpp::String() calls in the code, but (hopefully) these are not called hundreds of times (unless when parsing the shared string table) and would require further testing in non UTF-8 operating system and R. And I don't have any non UTF-8 system and I don't want to further research this :)