Fix standardise_taxon_rank() mangling 'section'

[ehwenk]

"sectio" is a literal prefix of its own English translation
("section"), so a fixed substring replacement anywhere in the
input also matched (and mangled) already-English values:
"section"/"subsection"/"zoosection"/"zoosubsection" became
"sectionn"/"subsectionn"/etc.

Anchor the replacement to the end of the string instead --
generalises to any "...section" variant without an allowlist.

Found via real AFD data in traitecoevo/taxonAlign#11.

[ehwenk]

Root cause

Of the seven Latin→English rank translations in standardise_taxon_rank(), "sectio" → "section" is the only pair where the find-string is a literal prefix of its own replace-string. That's what caused the bug: a fixed-string substring replacement (gsub(fixed = TRUE)) matches "sectio" anywhere in the input, including inside values that are already English -- "section", "subsection", "zoosection", "zoosubsection" all contain "sectio" as a substring, so the replacement fired again and left the original trailing "n": "sectionn", "subsectionn", etc.

None of the other six pairs (regnum/kingdom, classis/class, ordo/order, familia/family, varietas/variety, forma/form) have this property -- checked each; none is a prefix of its own translation.

Fix and rejected alternative

Anchored the "sectio" replacement to the end of the string (gsub("sectio$", "section", x = .)) rather than matching anywhere. A genuine Latin value ("sectio", "subsectio", ...) always ends in "sectio"; an already-English value ends in "ection", not "sectio" -- so this distinguishes the two cases correctly.

Considered an alternative: skip translation for any input already in an allowlist of known-English terms. Rejected because it would need every current and future "...section" compound spelled out to stay correct. Confirmed the anchored fix instead handles an untested compound for free:

standardise_taxon_rank("supersectio")
#> [1] "supersection"

Verification

standardise_taxon_rank(c("section", "sectio", "subsection", "subsectio",
                          "zoosection", "zoosectio", "zoosubsection", "zoosubsectio"))
#> [1] "section"       "section"       "subsection"    "subsection"
#> [5] "zoosection"    "zoosection"    "zoosubsection" "zoosubsection"

Added a regression test to tests/testthat/test-functions-standardise_names.R that deliberately interleaves already-English and still-Latin values in one input vector (rather than testing each kind in isolation), matching how real data actually looks -- a mix, not neatly pre-sorted batches. testthat::test_file() on that file: 7/7 passing, 0 failures.

Found while building an APCalign-equivalence test suite in the sibling taxonAlign package -- see traitecoevo/taxonAlign#11 for the original report (411/314218 rows affected in a real AFD-derived reference table).