Correct unsupported v21.0 default seam metadata

[farach]

Scientific correction

Independent reconstruction in farach/onet-vintage-bias established that v21.0 is not proven as a global O*NET content/method seam. The package's prior claim that v21.0 "retired the Task Relevance scale" was unsupported. This PR removes v21.0 as a package-verified default seam while preserving the verified v25.1 SOC-2010 to SOC-2018 taxonomy seam and the existing seams = override API.

Changes

• R/resurvey.R: onet_known_seams() now returns only the verified v25.1 soc_seam row (2020-11-01). The scale_seam / v21.0 / 2016-08-01 row is removed.
• Default v20.1 -> v21.x comparisons with unchanged SOC vintage are no longer seam-flagged solely for crossing that release date; ordinary non-seam comparability rules apply.
• The verified v25.1 taxonomy seam behavior is unchanged (still flagged, still safely_comparable = FALSE).
• The seams = override API on onet_content_change() and onet_resurvey_panel() is unchanged. A caller with channel-specific evidence can still supply a custom v21.0 row; documentation now makes explicit that such custom seams require external justification and are not package-verified defaults.
• Updated roxygen docs in R/resurvey.R and R/content_change.R, regenerated man/onet_content_change.Rd and man/onet_resurvey_panel.Rd via devtools::document().
• Corrected NEWS.md, docs/SPEC.md, and docs/DATA_NOTES.md prose that asserted v21.0 as a verified/default seam or claimed retirement of the Task Relevance scale. Added a NEWS bug-fix entry describing this as removal of unsupported default metadata (not a new scientific finding).
• Regenerated tests/testthat/fixtures/content-change-valid.rds to match the corrected default output.
• Added/updated tests (test-first): default seam registry contains only v25.1; default v20.1->v21.x pair is unflagged; caller-supplied v21.0 seam still flags the comparison; v25.1 seam remains flagged.

Scope search

Searched the entire tracked tree for v21, 21.0, 2016-08-01, relevance seam, retired Task Relevance, post-v21, and safely comparable. Confirmed matches in README.md/README.Rmd, the within-between-decomposition vignette, measure-reproducibility vignette, R/measure.R, and man/onet_task_to_occupation.Rd are unrelated (generic "task relevance" scale references or an unrelated safely_comparable usage in onet_panel_reconcile()), so they were left untouched.

Validation

• Red evidence: new/updated tests failed against unmodified main (3 failures in test-content-change.R, 8+ in test-resurvey.R before implementation).
• Green evidence: after implementation, devtools::test() full suite passes: [ FAIL 0 | WARN 0 | SKIP 0 | PASS 942 ].
• devtools::document() run; Rd files regenerated and match roxygen source.
• rcmdcheck::rcmdcheck(args = "--as-cran"): 1 ERROR | 1 WARNING | 1 NOTE, all from a pre-existing quarto -V invocation quirk in checking DESCRIPTION meta-information (a TMPDIR= argument is misparsed as a subcommand on this Windows box) and a New submission NOTE. Confirmed identical result on unmodified main (verified via git stash), so this is a pre-existing environment issue, not introduced by this change.
• Diff scope is exactly 10 files: NEWS.md, R/content_change.R, R/resurvey.R, docs/DATA_NOTES.md, docs/SPEC.md, man/onet_content_change.Rd, man/onet_resurvey_panel.Rd, tests/testthat/fixtures/content-change-valid.rds, tests/testthat/test-content-change.R, tests/testthat/test-resurvey.R. No unrelated API/output change. No credentials or cache data touched. Worktree is clean after commit.

Handoff

• Base SHA: 353f7cb8d0dd3fc35ff894caab52245ee3048a26 (verified against remote main before editing)
• Head SHA: 761e77420ea0896b8d669d78926f2f58a3da708d
• Branch: farach-correct-v21-seam-metadata

Stopping here for independent review per instructions; not merging, tagging, releasing, or deleting branches.

Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com