Fix @returns alias (#24) and unquoted-name false positive (#23)

[TroyHernandez]

Two issues reported by @grantmcdermott while running tinyrox on tinyplot.

#24 — @returns rejected as unknown tag

roxygen2 supports @returns as an alias of @return; tinyrox only allowlisted @return, so a single @returns aborted document(). Added @returns to the tag allowlist and routed it to the same Rd slot.

#23 — dependency package name flagged in Description prose

check_description_cran() cross-referenced the package's own dependency names against the Title/Description prose, so "base R graphics system" got flagged because graphics is an import (and fix = TRUE rewrote it to 'graphics'). CRAN only wants quoting when you refer to a package/software by name, not for ordinary words.

The check is now tiered by confidence:

• Curated software names (OpenAI, TensorFlow, ...) — unambiguous proper nouns, so a directive warning and auto-quoting by fix = TRUE / fix_description_cran().
• Dependency names — ambiguous in prose, so an advisory warning only, and never rewritten by the auto-fix paths.

So graphics is flagged with a soft "ordinary-word uses are fine" note and left untouched, while OpenAI still gets the hard nudge.

Version bumped to 0.3.3.6, tinytests added for both, R CMD check --as-cran clean (0/0/0).

Closes #23
Closes #24

[TroyHernandez]

Thanks @grantmcdermott, both fixed here (v0.3.3.6).

@returns is now a first-class alias of @return, so it won't abort document() anymore (#24).

The graphics one was a genuine false positive (#23). The check was matching your own dependency names against the Description prose, so "base R graphics system" tripped on the graphics import, and fix = TRUE would even rewrite it to 'graphics'. CRAN only wants quoting when you're naming the software, not for ordinary words, so dependency names now get a soft advisory and the auto-fix leaves them alone. Curated names like OpenAI still get the firm nudge since there's no ordinary-word use of those.

Appreciate you putting tinyrox through its paces on tinyplot.

[TroyHernandez]

Quick follow-up, @grantmcdermott: digging into your two reports made me question two of tinyrox's own defaults, and I'm planning to fold the answers into this PR rather than spin up separate ones.

1. Unknown tags currently hard-error and abort the whole document() run, which is why a single @returns blocked you entirely. I'm leaning toward warn-and-skip instead (roxygen2's behavior), so one unlisted tag can't take down the build.

2. The Description unquoted-name check that flagged graphics has no roxygen2 or R CMD check equivalent and leans on a hardcoded "known software" list. I'm reconsidering whether a minimal doc generator should be in that business at all.

Both would land as amendments to this PR, not blockers. Since you started me down this path, curious whether you'd call the strict-allowlist hard error a feature or a footgun.

[grantmcdermott]

Thanks Troy. Both sound like good ideas to me. Quick responses/thoughts:

1. Yes, a warning seems a better default than erroring out.
2. Same. The current approach is probably fragile (as we've seen) and I'm not sure that it buys you all that much that in the first place.;R CMD check will already error if your DESCRIPTION is misconfigured.