pars(terms =) now reports as defunct for every input type

[joethorley]

Closes #108

Only pars.term() carried terms as a formal, so every other method absorbed it into ... and failed with chk_unused()'s message, which named neither the defunct argument nor its replacement:

pars(term("a[1]"), terms = TRUE)
#> Error: The `terms` argument of `pars()` was deprecated in term 0.2.0 and is now defunct.

pars(c("a[1]"), terms = TRUE)
#> Error: `...` must be unused.

0.4.0 escalated this argument from a warning to a hard error, so this is precisely when people are migrating call sites. Character input is the likely case in practice, since pars() accepts it directly.

Approach

stop_defunct_terms() checks ...names() for terms and raises the same deprecate_stop() error. It is called before chk_unused() in pars.default(), pars.character() and pars.term_rcrd(); pars.term() is unchanged, as its formal already handles it.

Intercepting ... rather than adding a terms formal to the other three methods keeps the defunct argument out of their signatures and so out of their \usage sections.

...names() needs R >= 4.1, which DESCRIPTION already requires.

Verification

All five input paths now give the same message:

call	before	after
pars(term("a[1]"), terms = TRUE)	defunct error	defunct error
pars(term("a[1]"), terms = FALSE)	defunct error	defunct error
pars(c("a[1]"), terms = TRUE)	... must be unused	defunct error
pars(as_term_rcrd(...), terms = TRUE)	... must be unused	defunct error
pars(matrix(1:4, nrow = 2), terms = TRUE)	... must be unused	defunct error

The change is deliberately narrow, and a second test pins that down: genuinely unused arguments still report as unused.

pars(c("a[1]"), nope = TRUE)
#> Error in `pars.character()`: `...` must be unused.

devtools::test() 574 pass, 0 fail. devtools::check() 0 errors, 0 warnings, 0 notes.

Notes

NEWS.md is fledge-generated, so it is not edited here; the commit subject is written to serve as the release note.

Out of scope, but worth a look separately: npdims() had its own terms argument retired in 0.3.0 and carries no equivalent handling, so it may have the same gap.