fix(nitrogen): port sjos_levels in the afsetools order it documents

[lbm364dl]

What was wrong (mechanism)

whep::sjos_levels documents its level labels and colours as ported by value
from afsetools::load_vectors(), but afsetools wraps SJOS_levels in rev():

# afsetools/R/load_vectors.R:684
assign("SJOS_levels", rev(c("Within_boundary Under", ..., "Exceedance Over")), envir = env)

so its realised element 1 is "Exceedance Over". whep assigned order = 1L to
"Within_boundary Under" — the ordering was inverted relative to the vector it
says it ported. Verified by reading both files; the six colours match
SJOS_colours exactly, only the order differed.

No classification value changes: classify_sjos_n() uses sjos_levels$level
only as the factor levels of sjos_class, so membership is unaffected. What
changes is levels(sjos_class), i.e. the stacking and legend order of any
figure built from these levels — and Global stacks on the reversed vector
(Global/R/SJOS_N_figs.R:188, Global/R/SJOS_N.R:241), so the reversed order
is the one that reproduces its figures.

Classification: mechanical (faithfulness to a named source), with one visible
presentation consequence stated above. The issue offered the alternative of
keeping whep's ascending order and rewording the @source instead; I took the
faithful-port option because nourish_levels is already literal about its own
source vector (Nour_levels, unreversed at load_vectors.R:683), so the two
tables now port from afsetools the same way, and because it is the order
Global's figures use. Happy to flip to the doc-only option if you prefer the
ascending reading — say so and I will.

What changed

• data-raw/sjos_n_coefficients.R: sjos_levels rows reversed
  ("Exceedance Over" = 1L … "Within_boundary Under" = 6L), with a comment
  naming load_vectors.R:684 and the rev().
• data/sjos_levels.rda rebuilt by re-running the builder (the other three
  tables it writes came out byte-identical).
• R/datasets_sjos_n.R: @description/@format/@source now state the
  ordering explicitly, so it is a stated ported choice rather than an accident.
• tests/testthat/test_datasets_sjos_n.R: the order assertion now compares
  against rev(<ascending vector>) — written the way afsetools writes it — and
  additionally asserts order == seq_along(level).

Verification

man/ regenerated with devtools::document(), air format . run, both
changed R files and the data-raw script lint clean.

Test failing before the fix (old data/sjos_levels.rda restored via
git stash, new test in place):

Failure ('test_datasets_sjos_n.R:47:3'): sjos/nourish level tables are ordered and coloured
s$level (`actual`) not equal to `afsetools_sjos_levels` (`expected`).

    actual                     | expected
[1] "Within_boundary Under"    - "Exceedance Over"          [1]
[2] "Within_boundary Adequate" - "Exceedance Adequate"      [2]
[3] "Within_boundary Over"     - "Exceedance Under"         [3]
[4] "Exceedance Under"         - "Within_boundary Over"     [4]
[5] "Exceedance Adequate"      - "Within_boundary Adequate" [5]
[6] "Exceedance Over"          - "Within_boundary Under"    [6]

[ FAIL 1 | WARN 0 | SKIP 0 | PASS 27 ]

Passing after, together with the consumers and the data-raw freshness gate:

devtools::test(filter = "datasets_sjos_n|sjos_n_class|data_raw_freshness")
[ FAIL 0 | WARN 0 | SKIP 0 | PASS 105 ]

Not carried (out of scope, still open in the issue text): afsetools
Nourish_colours_tot / Nourish_levels_tot, the "Total" variants.

Closes #754.

🤖 Generated with Claude Code